Chapter 2: Survival Models and Mortality

1. Models for Survival and Mortality

The Survival Function:

▶ A standard way of describing survival and mortality is by using a
random variable for the time from birth until death. The age at
death is a random variable and is denoted by T0 (alternative
notations are X or T (0)).

▶ T0 is a continuous, positive random variable (T0 > 0). The

distribution function of T0 is F0 (t) = P[T0 ≤ t] = probability that
the new born dies before or on attaining age t. F0 (t) is also denoted
by FT0 (t).

▶ The survival functionor survival distribution function of T0 is

denoted by S0 (t) and S0 (t) = 1 − F0 (t) = probability that the new
born dies after attaining age t= the probability that the newborn
survives to at least age t.
1. Models for Survival and Mortality

The Survival Function:

▶ Some basic assumptions: (i) F0 (0) = 0 and S0 (0) = 1 (the newborn
is alive with probability 1). (ii) limt→∞ F0 (t) = 1 and
limt→∞ S0 (t) = 0 (the probability of the newborn surviving forever
is 0). (iii) F0 (t) is non-decreasing and S0 (t) is non-increasing (as x
increases, there is an increasing probability that the newborn dies by
age t).

▶ The upper age limit on a survival distribution is denoted by ω which

is usually a finite number such as 100 or 110 for human survival. If
ω = 100, then F0 (t) = 1 and S0 (t) = 0 for any t ≥ 100.

Time Until Death for (x): The notation (x) refers to an individual alive
at age x. Tx denotes the continuous random variable measuring the time
until (x)’s death from age x. Note that given that a newborn survives to
age x, the future lifetime measured from age x is Tx = T0 − x.
1. Models for Survival and Mortality
▶ Tx can be regarded as the conditional random variable of additional
number of years to be lived after age x given that the individual has
survived to age x. When we are measuring Tx , it is understood that
the person is alive at age x, so the distribution of Tx is represented
as conditional distribution of T0 :

P[Tx ≤ t] = P[T0 − x ≤ t|T0 > x] = P[T0 ≤ x + t|T0 > x]

P[x < T0 ≤ x + t] F0 (x + t) − F0 (x) S0 (x) − S0 (x + t)
= = = .
P[T0 > x] 1 − F0 (x) S0 (x)
▶ The distribution function of Tx is denoted by Fx (t) = P[Tx ≤ t] and
it is denoted by t qx = Fx (t) = probability that (x) will die within
the next t years (by age x + t).

▶ The survival function of Tx is Sx (t) which is denoted by t px and it

is probability that (x) survives at least to time t (and dies some time
after t).
1. Models for Survival and Mortality
▶ By the conventions of actuarial notation, for mortality or survival for
a 1 year period, we write 1 qx = qx and 1 px = px . It is assumed that
∞ qx = 1 (the probability of eventually dying is 1) and ∞ px = 0 (the
probability of living forever is 0). It is also assumed that 0 qx = 0
and 0 px = 1.

Factorization of Survival Probability and Deferred Mortality

Probability: For n ≤ t, the probability that (x) survives at least t-years
can be factored in the following way:

t px = n px · t−n px+n .

The interpretation of this factorization is that in order for (x) to survive t

years, (x) must survive the first n years (event A, probability n px ) and
then must survive t − n more years from age x + n (event B|A,
probability t−n px+n ). The probability of surviving t years is the event
A ∩ B, and we use the rule that P[A ∩ B] = P[A]P[B|A]. As a special
case, x+n p0 = x p0 · n px and it follows that n px = S0S(x+n)
0 (x)
= Sx (n).
1. Models for Survival and Mortality
▶ If t is an integer, the survival probability over the t-year can get
factorization form t px = px · px+1 · · · px+t−1 .

▶ An important probability in life contingencies is the deferred

mortality probability. The probability that (x) survives t years and
then dies in the u years following that is

t|u qx = t px · u qx+t = t+u qx − t qx = t px − t+u px

S0 (x + t) − S0 (x + t + u)
= P[t < Tx ≤ t + u] =
S0 (x)
= Sx (t) − Sx (t + u).

This is a t-year deferred, u year mortality probability; it is the

probability that (x) survives t years to age x + t and then dies by
age x + t + u. Note that this is not simply equal to u qx+t , since (x)
must first survive to age x + t and then from age x + t must die in
the following u years. By notational convention, we write
t|1 qx = t| qx . Also, we note that 0| qx = qx and t+u qx = t qx + t|u qx .
1. Models for Survival and Mortality

▶ Example: For the survival functions S0 (x) = e −ax , where a > 0 and
x 2


S0 (x) = 1 − 100 for 0 ≤ x ≤ 100. Find F0 (x), t py and t|u qy .

Solution: For S0 (x) = e −ax , we have that F0 (x) = 1 − e −ax , and

e −a(y +t) −at
t py = e −ay = e . We also have t|u qy = e −at − e −a(t+u) .
x 2 x 2



For S0 (x) = 1 − 100 , we have F0 (x) = 1 − 1 − 100 , and
 2
(100−y −t)2 /1002 t
t py = (100−y )2 /1002 = 1 − 100−y . We also have

 2  2
t t +u
t|u qy = 1− − 1− .
100 − y 100 − y
1. Models for Survival and Mortality
▶ Curtate Future Lifetime of (x): For someone alive at age (x), Kx
denotes the completed (integer) number of years until (x)’s death
(doesn’t count the fractional part of the year in which death occurs).
Kx is a non-negative discrete integer-valued random variable.
Kx = 0 corresponds to (x) not surviving 1 complete year (dying
before age x + 1), which has probability
P[Kx = 0] = P[0 < Tx ≤ 1] = qx = 0| qx .

▶ Kx = 1 denotes the event that (x) survives 1 complete year and dies
during the second year (between ages x + 1 and x + 2) and the
probability is P[Kx = 1] = P[1 < Tx ≤ 2] = 1|1 qx = 1| qx . For the
non-negative integer k, Kx = k corresponds to (x) surviving k
complete years and dying during the k + 1-st year, which has

P[Kx = k] = P[k < Tx ≤ k + 1] = k|1 qx = k| qx

=k px · qx+k = k px − k+1 px = k+1 qx − k qx .
1. Models for Survival and Mortality
▶ Note that P[Kx ≤ k] = P[Tx ≤ k + 1], surviving at most k complete
years means that death occurred before the completion of k + 1
years. The notation Kx can be simplified to K if the age x is clear
and obvious.

▶ If n ≥ t are integers, then

n−1
X
(i) n qx = t qx + t|n−t qx = qx + 1| qx + · · · + n−1| qx = k| qx
k=0

and as a special case,

∞
X
(ii) 1 = ∞ qx = qx + 1| qx + 2| qx + · · · = k| qx .
k=0

The interpretation of (i) is that in order for (x) to die within n

years, death must occur in either the 1st year, the second year, . . .,
or the n-th year. The interpretation of (ii) is that (x) must
eventually die in some future year with probability 1.
1. Models for Survival and Mortality
▶ Example: You are given the survival function S0 (x) = (1+x) 1
2 for

x ≥ 0. Calculate:
(i)10 q10 , (ii)10|10 q10 , (iii)t px , (iv )t|u qx , (v )P[K10 = 10],
(vi)P[Kx = k].
2
Solution: We have 10 q10 = S0 (10)−S 0 (20)
S0 (10) = 1 − 1/21
1/112 = 0.7256 and
2
S0 (20)−S0 (30) −1/312
10|10 q10 = S0 (10) = 1/211/11 2 = 0.1485. We also can get
S0 (x+t) 1/(x+t+1)2 (x+1)2
t px = S0 (x) = 1/(x+1)2 = (x+t+1)2 and
S0 (x+t)−S0 (x+t+u) (x+1)2 (x+1)2
t|u qx = S0 (x) = (x+t+1) 2 − (x+t+u+1)2 ,

P[K10 = 10] = 10| q10 = S0 (20)−S S0 (10)

0 (21)
= 0.0244,
(x+1)2 (x+1)2
P[Kx = k] = k| qx = (x+k+1)2 − (x+k+2)2 .

▶ Example: Show that P[T20 > 10|T20 > 5] = 5 p25 .

Solution:
P(T20 >10) 10 p20 5 p20 5 p25
P[T20 > 10|T20 > 5] = P(T20 >5) = = = 5 p25 .
5 p20 5 p20
2. The Force of Mortality
▶ Force of Mortality and Probability Density Function: The
probability density function of a continuous random variable Y ,
denoted by f (y ) or fY (y ) is the derivative of the distribution
function fY (y ) = FY′ (y ). The pdf of T0 can be formulated as

d d
f0 (x) = F0′ (x) = −S0′ (x) = q = − x p0 .
dx x 0 dx
▶ For a continuous random variable Y , the hazard rate or failure rate
at t is defined to be
fY (t) fY (t) −SY′ (t) −d
hY (t) = = = = ln[SY (t)].
1 − FY (t) SY (t) SY (t) dt

For the time until death random variable T0 , the hazard rate is
called the force of mortality. The µx notation is standard actuarial
notation for force of mortality:
d d
f0 (x) S ′ (x) d p dx x q0
µx = =− 0 = − ln[S0 (x)] = − dx x 0 = .
1 − F0 (x) S0 (x) dx p
x 0 x p0
2. The Force of Mortality

▶ µx ≥ 0 for all x and µx can be interpreted as the instantaneous rate

of mortality at age x given alive at age x, measured on an
annualized basis. We therefore have probability density function of
X in terms of the force of mortality

f0 (x) = S0 (x)µx = x p0 · µx .
d
From the relationship µx = − dx ln[S0 (x)], it is not difficult to derive
that Rx
S0 (x) = x p0 = e − 0 µt dt .
▶ Example: For each of the following forces of mortality, find S0 (y ),
10 q10 and t py , where

1
(i)µx = for x ≥ 0, (ii)µx = c for x ≥ 0.
1+x
2. The Force of Mortality

Ry 1
▶ Solution: (i) We have S0 (y ) = e − 0 1+x
dx
= e − ln(1+y ) = 1
1+y . Also,
S0 (20) 1/21 10
we have 10 q10 = 1 − 10 p10 = 1 − S0 (10) =1− 1/11 = 21 and
S0 (y +t) 1/1+y +t 1+y
t py = S0 (y ) = 1/1+y = 1+y +t .
Ry
(ii) S0 (y ) = e − 0 cdx = e −cy . We also have
S0 (20) e −20c −10c
10 q10 = 1 − 10 p10 = 1 − S0 (10) = 1 − e −10c = 1 − e and
S0 (y +t) e −c(y +t)
t py = S0 (y ) = e −cy = e −ct .
2. The Force of Mortality

▶ Mortality probability can be expressed as the integral of the density

function.
Z x Z x
x q0 = F0 (x) = f0 (s)ds = s p0 µs ds.
0 0

▶ Recall that Tx is also a continuous random variable and the

probability density function of Tx is
d d d d
fx (t) = Fx (t) = t qx = − t px = − Sx (t).
dx dt dt dt
▶ The force of mortality t years after age x (age x + t) can be
formulated as
d
fx (t) p S ′ (t) S ′ (x + t) d
µx+t = = − dt t x = − x =− 0 = − ln S0 (x+t).
1 − Fx (t) t px Sx (t) S0 (x + t) dt
2. The Force of Mortality
▶ A survival distribution in which the force of mortality depends only
on the age of the individual is referred to the aggregate survival
distribution. There is no difference between µ60 and µ40+20 and
µ5+55 . Also, note that µx+t can be described as the hazard rate
hTx (t) for random variable Tx .

▶ The pdf of Tx is fx (t) = d d d d

dt Fx (t) = − dt Sx (t) = dt t qx = − dt t px =
S ′ (x+t)
t px · µx+t = − 0S0 (x) .

▶ Some important relationships involving Tx are

R x+n Rn
(i) ln(n px ) = − x µs ds =R − 0 µx+t dt.R
n n
(ii) n qx = P[0 < Tx ≤ n] = 0 fx (s)ds = 0 s px · µx+s ds and
R1
qx = 0 s px · µx+s ds
R x+n Rn
(iii) n px = e − x µs ds =Re − 0 µx+t dt , R
∞ ∞
n px = P[Tx > n] = n fx (s)ds = n s px · µx+s ds and
R t+u
t|u qx = t s px · µx+s ds.
2. The Force of Mortality

▶ Some important relationships involving Tx (cont.)

R∞
▶ As ∞ px = e − 0 µx+s ds = 0, it follows that ∞ µx+s ds = ∞.
R
R∞ 0
▶ as a pdf, fx (s) = s px µx+s must satisfy p µ ds = 1.
0 s x x+s
▶ d t px = −t px · µx+t and d t px = t px · [µx − µx+t ].
dt dx

▶ Note that with an upper age limit of ω in a survival distribution, for

someone currently at age x, the term ∞ refers to ω − x. For
instance, if ω = 110 and x = 40, then ∞ = 70 = 110 − 40.

▶ The way in which we will see mortality distributions presented will

most likely be one of the following ways:
▶ we are given a general form for S0 (y ), the survival function; we can
find t px = S0S(x+t)
0 (x)
, and we can find µy , fx (t), etc.
▶ we are given a general form for µy , the force of mortality; we can
Rt
find t px = e − 0 µx+s ds and we can find S0 (y ) = y p0 , etc.
2. The Force of Mortality

▶ Example: You are given (i) µx = A + e x and (ii)0.50 p0 = 0.5.

Calculate A.
R 0.5 x 0.5
Solution: We have 0.5 = e − 0 (A+e )dx = e −0.5A e −(e −1) , it
follows that ln(0.5) = −0.5A − (e 0.5 − 1) so A = 0.089.

▶ Example: The force of mortality is µy = 1+y1

. Find the median
future lifetime of a newborn, and find the median future lifetime of
someone alive at age 10.
Rx
Solution: We have S0 (x) = e − 0 µy dy = 1+x
1
. The median future
lifetime of a newborn is a constant such that S0 (m) = 12 . So we get
1 1
1+m = 2 and m = 1.
The median future lifetime of (10) is m such that m q10 = 12 . So, we
get m p10 = S0S(10+m)
0 (10)
= 1/1+10+m 11 1
1/1+10 = 11+m = 2 . We have m = 11.
2. The Force of Mortality

▶ Example: Suppose that S0 (x0 ) = 0.5. Show that

fx0 (t) = 2f0 (x0 + t).

Solution: We can compute that

x0 +t p0
fx0 (t) = t px0 · µx0 +t = µx0 +t = 2x0 +t p0 µx0 +t = 2f0 (x0 + t).
x0 p0
3. The Life Table

Life Table Notation

▶ With a starting group of l0 newborns, lx denotes the expected
number of survivors at age x from the original group. Some
definitions and relationships involving life table functions are:
(i) dx = lx − lx+1 is the number of deaths between ages x and x + 1.
(ii) lx+1 = lx − dx .
(iii) The number of deaths between ages x and x + n is
n dx = lx − lx+n = dx + dx+1 + · · · + dx+n−1 .
d
and n qx = nlxx = lx −llxx+n .
l
(iv) px = x+1 lx
, qx = dlxx , n px = lx+n
lx
dx+k
(v) k| qx = lx (of the lx people who are alive at age x, dx+k of them
survive to age x + k and die before attaining age x + k + 1, so the
proportion of the lx who die between ages x + k and x + k + 1 is
k| qx . )
d
n|m qx = lx+n −llxx+n+m = m lx+n
x
where m dx+n is the number of deaths
occurring in the life table between ages x + n and x + n + m.
l
(vi) S0 (y ) = ly0 .
Rn
(vii) lx − lx+n = n dx = 0 lx+t · µx+t dt
3. The Life Table

Life Table Notation

(cont.)
d
Rn ly
(viii) lx+n = lx · e − 0
µx+t dt
and µy = − dyly .
d d
(ix) dt t px = −t px · µx+t , dx t px = t px · [µx − µx+t ].
2
d d d
(x) dy ly = −ly · µy , dy 2 ly = − dy [ly · µy ].

Example 1: A life table has the following form: lx = (1.01)−x · l0 . Find

(i) qx ; (ii) k| qx ; (iii) µx .
lx+1 (1.01)−x−1
Solution: We have qx = 1 − px = 1 − lx =1− (1.01)−x = 0.0099.
−x−k
lx+k −lx+k+1 −(1.01)−x−k−1
k| qx = lx = (1.01) (1.01) −x = (0.01)(1.01) −k−1
. We also
d
− dx lx (1.01)−x ln( 1.01
1
)
have µx = lx = − (1.01)−x = ln(1.01) = 0.00995.
3. The Life Table

▶ Example 2: You are given the following information about a life

table. l0 = 100, l2 = 90, l4 = 75, l6 = 50, l8 = 15 and l10 = 0. Also,
lx is a linear function on each interval [0, 2], [2, 4], etc. Find each of
the following µ5 , µ5.5 , 2 p5 and 2 p5.5 .

Solution: We have l4+t = 75 − 12.5t for 0 ≤ t ≤ 2.

µ4+t = − dl4+t /dt
l4+t
−12.5
= − 75−12.5t so µ5 = 0.2 and µ5.5 = 0.22. We also
have l6+t = 50 − 17.5t for 0 ≤ t ≤ 2. 2 p5 = ll57 = 75−12.5
50−17.5
= 0.52
l7.5 50−(17.5)(1.5)
and 2 p5.5 = l5.5 = 75−(12.5)(1.5) = 0.422.
4. Mean and Variance of Tx and Kx
Expected Value and Variance of Tx
▶ The expected value of the random variable Tx is referred to the
complete expectation of life for (x) and we have
Z ∞ Z ∞
E[Tx ] = e̊x = t · t px µx+t dt = t px dt.
0 0

▶ Note that ∞ = ω − x if the upper age limit for survival is a finite

age ω. Since we are assuming that the individual is alive at age x
and we are measuring Tx = T0 − x which is the time until death
from age x, e̊x can also be thought of as a conditional expectation
e̊x = E[T0 − x|T0 > x]. This is also called the mean residual lifetime
given alive at age x.

▶ The variance of Tx is

Var[Tx ] = E[Tx2 ] − (E[Tx ])2 = E[Tx2 ] − (e̊x )2 ,

R∞ R∞
where E[Tx2 ] = 0 t 2 · t px · µx+t dt = 0 2t · t px dt.
4. Mean and Variance of Tx and Kx
▶ The n-year term expectation of life for (x) is
Z n Z n
e̊x:n̄| = t px dt = n · n px + t · t px · µx+t dt
0 0

where e̊x:n̄| represents the average number of years lived by (x)

between ages x and x + n. If we define the random variable
Tx,n = min{Tx , n}, then e̊x:n̄| = E[Tx,n ]. In loss distribution context,
this is the limited expected value of Tx , with limit n. Note that
R1
e̊x = e̊x:n̄| + n px e̊x+n and e̊x = 0 t px dt + px e̊x+1 .

▶ The median future lifetime of (x) is m(x), where m(x) is the solution
of the equation m(x) px = P[Tx > m(x)] = 0.5, or equivalently in the
life table, lx+m(x) = 12 lx . m(x) is the number of years, as measured
from age x, at which half of those at age x are still alive. If
0 ≤ k ≤ 100, the k-th percentile of Tx is the point ck for which
ck px = P[Tx > ck ] = 1 − 0.01k. The median is the 50th percentile.
4. Mean and Variance of Tx and Kx

▶ The mode of Tx is the point t at which the pdf fx (t) is maximized.

2
x
▶ Example: Using S0 (x) = 1 − 100 for 0 ≤ x ≤ 10 = ω, find the
mean, variance, median and mode of T4 .
R 10−4 R 6 84−8t−t 2
Solution: We have e̊4 = 0 t p4 dt = 0 84 dt = 3.43.
2
R6 R6 84−8t−t 2
E[T4 ] = 0 2t t p4 dt = 0 2t( 84 )dt = 14.57 and hence
Var[T4 ] = 2.81. The median of T4 is the solution to
84−8m(4)−m2 (4)
m(4) p4 = 84 = 0.5 and hence m(4) = 3.62. The mode of
T4 occurs at the value of t that maximizes f4 (t) = 8+2t 84 with
0 ≤ t ≤ 6 and we find t ∗ = 6.
4. Mean and Variance of Tx and Kx
Curtate Expectation of Life for (x)
▶ The expected value of the discrete random variable Kx is called the
curtate expectation of life for (x) and is denoted by ex . We have
∞
X ∞
X
E[Kx ] = ex = k·P[Kx = k] = k·k| qx = 0·0| qx +1·1| qx +2·2| qx +· · · .
k=0 k=0

The final expression can be rewritten in the form

∞
X
(px − 2 px ) + 2(2 px − 3 px ) + 3(3 px − 4 px ) + · · · = k px .
k=1

Note that ex ≤ e̊x and if the survival distribution has a finite integer
upper limit ω, the maximum integer age that (x) can attain is
ω − 1, and the maximum number of complete years that (x) can
P∞ Pω−x−1
survive is ω − x − 1 and k=1 = k=1 .
P∞
▶ The key formulation for the curtate expectation is ex = k=1 k px .
4. Mean and Variance of Tx and Kx

▶ The n-year term curtate expectation is ex:n̄| = nk=1 k px . Note that

P
ex = ex:n̄| + n px ex+n and ex = px + px ex+1 = px (1 + ex+1 ). The
proof of ex = px + px ex+1 is based on the factorization relationship
t px = px · t−1 px+1 . We have

∞
X
ex = k px = px + 2 px + 3 px + · · · = px + px · px+1 + px · 2 px+1 + · · ·
k=1
= px + px (px+1 + 2 px+1 + · · · ) = px + px · ex+1 .

▶ Note also that Tx = Kx + Rx , where Rx denotes the fraction of the

year lived during the year of death, and then e̊x = ex + E[Rx ].
4. Mean and Variance of Tx and Kx

▶ The two main recursion relationship for life expectations are: (i)
ex = px + px ex+1 , for curtate expectation and (ii)
R1
e̊x = 0 t px dt + px e̊x+1 for complete expectation.

▶ Variations of the following relationship is very useful in the exam:

▶ ex:n̄| = px + px · ex+1:n−1| .
▶ ex:2̄| = px + px · ex+1:1̄| = px + px · px+1 = px + 2 px .
R1 R1
▶ e̊x:2̄| = e̊x:1̄| + px · e̊x+1:1̄| = 0 t px dt + px · 0 t px+1 dt.
▶ e̊x:3̄| = e̊x:1̄| + px · e̊x+1:2̄| = e̊x:1̄| + px · e̊x+1:1̄| + 2 px · e̊x+2:1̄|
R1 R 0.5
▶ e̊x:1.5| = e̊x:1̄| + px · e̊x+1:0.5| = 0 t px dt + px · 0 t px+1 dt.
2
▶ Example: The survival function is S0 (x) = 1 − 100x
for
0 ≤ x ≤ 10 = ω. (i) Find e4 . (ii) Suppose that as a result of a
change in living circumstances for year of age 4 to 5, the mortality
probability is doubled at that age only, but reverts back to the
original survival for x ≥ 5. Find e4 .
4. Mean and Variance of Tx and Kx

▶ Solution: (i) We have

e4 = p4 + 2 p4 + 3 p4 + 4 p4 + 5 p4 = 75+64+51+36+19
84 = 2.92.
(ii) We can use the relationship e4 = p4 (1 + e5 ). The original value
of p4 is p4 = SS00 (5) 75
(4) = 84 = 0.8929 and the original value of q4 is
0.10714. The revised value q4new = 0.2143 and p4new = 0.7857.
Because the mortality from age 5 is unchanged, we get
e5old = p5 + 2 p5 + 3 p5 + 4 p5 = 64+51+36+19
75 = 2.267. We get
e4new = p4new (1 + e5old ) = (0.7857)(1 + 2.267) = 2.567.
5. Parametric Survival Models
Some Parametric Survival Models
▶ The Uniform Distribution: This model is also referred to De Moivre’s
Law or the Linear Form for lx . Three definitions of the model are
▶ lx = k(ω − x) for 0 ≤ x ≤ ω
▶ S0 (x) = 1 − x for 0 ≤ x ≤ ω
ω
▶ µx = 1 for 0 ≤ x ≤ ω
ω−x
▶ Other important relationships are
▶ n px = 1 − n and n qx = n for 0 ≤ n ≤ ω − x
ω−x ω−x
▶ fx (t) = t px µx+t = 1 for 0 ≤ t ≤ ω − x
ω−x
▶ k| qx = 1 for 0 ≤ k ≤ ω − x − 1, t|u qx = u
ω−x ω−x
▶ Under this model, for any x between 0 and ω, Tx has a uniform
1
distribution on the interval [0, ω − x]. In this case, fx (t) = ω−x and
ω−x
the mean and variance of Tx are e̊x = E[Tx ] = 2 and
(ω−x)2
Var[Tx ] = 12
. We have
n n
n2
Z Z
lx+t 1 n
e̊x:n̄| = t px dt = dt = · [2(ω−x)−n] = n−
0 0 lx ω−x 2 2(ω − x)

The curtate expectation is ex = ω−x−12

. The name “De Moivre” is
used occasionally to indicate the linear model of mortality.
5. Parametric Survival Models

▶ Generalized DeMoivre’s Law: This survival model is described in the

equivalent ways:
▶ lx = k(ω − x)α for 0 ≤ x ≤ ω and α > 0
▶ S0 (x) = 1 − x α = ω−x α



ω ω
▶ µx = α
ω−x

▶ Other relationships for this generalization of DeMoivre’s Law are

 α
n
n px = 1 − ω−x so that Tx has a similar survival form to T0
except that Tx is defined on the interval [0, ω − x],
  α−1
α t ω−x
fx (t) = ω−x 1 − ω−x for 0 ≤ t ≤ ω − x, e̊x = α+1 and
2
α(ω−x)
Var[Tx ] = (α+1)2 (α+2) .
5. Parametric Survival Models
▶ Exponential Distribution: This distribution is referred to constant
force of mortality model. In this model, µx = µ for all x, t px = e −tµ
and px = e −µ for all x. The upper age limit is ω = ∞. The pdf of
Tx is fx (t) = t px µx+t = µe −tµ , so that Tx has an exponential
distribution with mean e̊x = µ1 and Var[Tx ] = µ12 . We also have
−kµ
k| qx = e (1 − e −µ ) and ex = e µ1−1 .

▶ The Gompertz Distribution: Under this model, the force of mortality

is µx = Bc x where B > 0 and c > 1. Then
S0 (x) = exp(− lnBc (c x − 1)) for y ≥ 0.

▶ The Makeham Distribution: Under this model, the force of mortality

is µx = A + Bc x where A > −B, B > 0, c > 1. Then
S0 (x) = exp(−Ax − lnBc (c x − 1)) for x ≥ 0.

n+1
▶ The Weibull Distribution: µx = kx n , S0 (x) = exp(− kxn+1 ) where
k > 0, n > 0 and x ≥ 0.
6. Fractional Age Assumption

▶ Uniform distribution of deaths (UDD) within each year of age: This

assumption is referred to the linear form of lx+t . It is also equivalent
to linear interpolation between lx and lx+1 for each integer age x.

▶ Under UDD assumptions, lx+t is a linear function for t on the

interval 0 ≤ t ≤ 1: lx+t = lx − t · dx for 0 ≤ t ≤ 1. Dividing both
sides of the equation by lx results in the relationship t px = 1 − t · qx ,
which can be written in the familiar UDD form t qx = t · qx for
0 ≤ t ≤ 1.

▶ Some important relationships:

▶ t qx = t · qx , fx (t) = t px µx+t = qx , µx+t = qx
1−t·qx
▶ t px+s = lx −(t+s)·dx = 1−(t+s)·qx if 0 ≤ s + t ≤ 1
lx −s·dx 1−s·qx
▶ t qx+s = t·qx if 0 ≤ s + t ≤ 1
1−s·qx
▶ e̊x = ex + 1 and e̊x:n̄| = ex:n̄| + 1 (1 − n px )
2 2
6. Fractional Age Assumption
▶ When dealing with fractional ages and UDD, it is usually most
convenient to represent events in terms of survival probabilities. For
instance, to find 0.25 qx+0.4 , we would use the relationships
lx+0.65
0.25 qx+0.4 = 1 − 0.25 px+4 = 1 − lx+0.4 . From UDD, we have
lx+0.4 = lx − 0.4dx and lx+0.65 = lx − 0.65dx . This can be extended
to periods beyond a current year of age. For instance,
lx+2+0.65 lx+2 −0.65dx+2
2.25 px+0.4 = lx+0.4 = lx −0.4dx .

▶ Recall that Kx is the curtate future lifetime of (x). Suppose that Rx

is the fraction of the year lived in the year of death when (x) dies.
Under UDD assumption, it is possible to show that Kx and Rx are
independent, and Rx has a uniform distribution on [0, 1]. As
Tx = Kx + Rx , it follows that under UDD assumption,
1
e̊x = E[Tx ] = E[Kx ] + E[Rx ] = ex + 21 and Var[Tx ] = Var[Kx ] + 12 .

▶ Constant force of mortality in each year of age: This assumption is

referred to the exponential form of lx+t . It is equivalent to
exponential interpolation between lx and lx+1 for each integer age x.
6. Fractional Age Assumption

▶ Under this model, lx+t has the form

 t
lx+1
lx+t = lx = (lx )1−t (lx+1 )t = lx (px )t
lx

for 0 ≤ t ≤ 1. It follows that that t px = (px )t for 0 ≤ t ≤ 1 and the

force of mortality is constant during each year of age, as
−d p d
µx+t = dtp t x = − dt ln(t px ) = − ln px for 0 < t < 1. This constant
t x
force is denoted by µx = − ln px .

▶ Some important relationships: px = e −µx , µx+t = − ln px if

0 < t < 1, t px = t px+s = e −tµx = (px )t if 0 ≤ t + s ≤ 1.

▶ Example: You are given q70 = 0.040 and q71 = 0.044. Calculate
e̊70:1.5| in the following two cases: (i) assume UDD in each year of
age (ii) assume constant force of mortality in each year of age.
6. Fractional Age Assumption

▶ Solution: (i) e̊70:1.5| = 01.5 t p70 dt = 01 t p70 dt + 11.5 t p70 dt. We

R R R
R1 R1 R1
know that 0 t p70 dt = 0 (1 − t q70 )dt = 0 (1 − t · q70 )dt =
R1
(1 − 0.04t)dt = 0.98 and
R01.5 R 1.5 R 1.5
1 t 70
p dt = 1 1 p70 t−1 p71 dt = (0.96) 1 t−1 p71 dt By changing
variable
R 1.5 s = t − R1 and ds = dt,R we have
0.5 0.5
p dt = 0 s p71 ds = 0 (1 − s q71 )ds =
R10.5 t−1 71 R 0.5
0
(1 − s · q71 )ds = 0 (1 − 0.044s)ds = 0.4945. Therefore, we
obtain that e̊70:1.5| = 0.98 + (0.96)(0.4945) = 1.455.
R1 R 1.5
(ii) Similarly, we have e̊70:1.5| = 0 t p70 dt + 1 1 p70 t−1 p71 dt =
R1 R 0.5
p dt + (0.96) 0 s p71 ds. Because of the constant force
0 t 70
assumption, we have t p70 = (p70 )t = (0.96)t and s p71 = (0.956)s
R1 R 0.5
and we get e̊70:1.5| = 0 (0.96)t dt + (0.96) 0 (0.956)s ds = 1.4545.