Survival distributions and life tables

Actuarial mathematics 3280

Department of Mathematics and Statistics
York University
Edward Furman
[email protected]

Edward Furman, Actuarial mathematics MATH3280 – p. 1/66

 Outline
After we have completed this theme you will be able to:
Define survival-time random variables (rv’s) for one life;
Calculate the expected values, variances, probabilities
and percentiles for survival-time rv’s;
Define the continuous type survival-time rv using a:
1. uniform distribution,
2. constant force of mortality,
3. hyperbolic assumption.
Be able to characterize life tables using the complete
expectation of life, median future lifetime, mode, etc.
Understand and apply the concept of select life tables.
Feel comfortable with the deterministic approach to life
tables.
Edward Furman, Actuarial mathematics MATH3280 – p. 2/66
 Survival function
A newborn’s child age-at-death, X , is a continuous non
negatively valued rv having cdf F (x) such that:
F (x) = P (X ≤ x), x ≥ 0.
Definition 1 We denote the survival function of X by S(x).
It is formulated as:
S(x) = P (X > x) = 1 − F (x),
and it is interpreted as the probability the newborn attains
age x.
We assume that F (0) = P (X ≤ 0) = 0 and thus S(0) = 1.
For X v F , we can produce, say the probability that the
newborn dies between ages x and z
P (x < X ≤ z) = F (z) − F (x) = S(x) − S(z), x < z.
Edward Furman, Actuarial mathematics MATH3280 – p. 3/66
Time-until death for a person aged x
The probability the newborn will die between ages x and z
given survival to age x is
F (z) − F (x) S(x) − S(z)
P (x < X ≤ z|X > x) = = .
1 − F (x) S(x)

The future lifetime of (x) that is X − x is denoted by T (x).

Thus, the probability that (x) dies between ages x and z is:
P (T (x) ≤ z − x).
Are the two latter equations equivalent?

Edward Furman, Actuarial mathematics MATH3280 – p. 4/66

Example 1 Suppose the cdf of (0) is given by
F (t) = t/80, 0 < t ≤ 80 and F (t) = 1 otherwise. What is the
probability that a new born:
1. dies no latter than age 30,
2. survives to age 50,
3. dies after age 30 but not latter than age 50.
Solution
(a) P (X ≤ 30) = F (30) = 3/8,
(b) P (X > 50) = 1 − F (50) = 1 − 5/8 = 3/8,
(c) P (30 < X ≤ 50) = F (50) − F (30) = 1/4.


Edward Furman, Actuarial mathematics MATH3280 – p. 5/66

To make probabilistic statements about T (x), we use:
1. Probability that (x) dies within t years or before x + t-th
birthday
t qx = P(T(x) ≤ t) this is the cdf of T(x);
if t = 1, then 1 qx = qx .
2. Probability that (x) attains age x + t
t px = P(T(x) > t) this is the survival function of T(x);
if t = 1, then 1 px = px .
3. Probability that (x) will die between ages x + t and
x+t+u
t|u qx = P(t < T(x) ≤ t + u) = t+u qx − t qx = t px − t+u px .

Edward Furman, Actuarial mathematics MATH3280 – p. 6/66

Assumption 1 The observation of survival at age x yields
the same conditional distribution of survival as the
hypothesis that a new born has survived to age x.
Then:
x+t p0
t px = P (T (x) > t) = P (X > x + t|X > x) = ,
x p0
x+t q0 − x q0
t qx = P (T (x) ≤ t) = 1 − P (T (x) > t) = .
x p0

Also, under Assumption 1, we have that:

t|u qx = t px · u qx+t .
Proof
t|u qx = P (t < T (x) ≤ t + u) = P (x + t < X ≤ x + t + u|X > x)

Edward Furman, Actuarial mathematics MATH3280 – p. 7/66

 Curtate future lifetime

P (X ≤ x + t + u) − P (X ≤ x + t)
t|u qx =
P (X > x)
S(x + t) − S(x + t + u)
=
S(x)
S(x + t) S(x + t) − S(x + t + u)
= ,
S(x) S(x + t)
as needed. 
Definition 2 Discrete rv K(x) denotes the number of future
years completed by (x) prior to death. K(x) is the greatest
integer in T (x) and its pmf is:
P (K(x) = k) = P (k ≤ T (x) < k + 1) = P (k < T (x) ≤ k + 1)
= k px · 1 qx+k = k|1 qx .

Edward Furman, Actuarial mathematics MATH3280 – p. 8/66

Note that k| qx is a pmf. Indeed:
∞
X ∞
X
P (K(x) = k) = k| qx = 1.
k=0 k=0
The cdf of K(x) is:
k
X k
X
P (K(x) ≤ k) = P (K(x) = h) = h| qx
h=0 h=0
Xk
= (h+1 qx − h qx ) = k+1 qx − 0 qx
h=0
= k+1 qx .

The cdf of K(x) is thus a step function:

FK(x) (y) = P (K(x) ≤ y) = k+1 qx , y ≥ 0,
Edward Furman, Actuarial mathematics MATH3280 – p. 9/66
and k is the greatest integer in y .
Example 2 Consider the cdf
r
x
F (x) = , 0 ≤ x ≤ 100.
100
What is the probability that a newborn lives at least 25
complete years?
Solution
P (K(0) ≥ 25) = 1 − P (K(0) < 25) = 1 − P (K(0) ≤ 24)
= 1 − 25 q0 = 1 − F(25) = 0.5.

Edward Furman, Actuarial mathematics MATH3280 – p. 10/66

 Force of mortality
Recall that the probability the newborn will die between
ages x and x + ∆x given survival to age x is
F (x + ∆x) − F (x)
P (x < X ≤ x + ∆x|X > x) = .
1 − F (x)
Then, taking lim∆x→0+ , we have that:
f (x)∆x
lim + P (x < X ≤ x + ∆x|X > x) = = µ(x)∆x ≥ 0,
∆x→0 S(x)
where
f (x) F 0 (x) −S 0 (x)
µ(x) = = =
S(x) S(x) S(x)
is called the force of mortality and for each age x, it gives
the value of the conditional density function of X at exact
age x given survival to that age.

Edward Furman, Actuarial mathematics MATH3280 – p. 11/66

The force of mortality can be used to specify the distribution
of X .
Proof Due to the definition of µ(y), we have that:
−S 0 (y) d
µ(y) = = − log S(y),
S(y) dy
which leads to
−µ(y)dy = d log S(y).
Integrating the above from x to x + n leads to:
Z x+n Z x+n
S(x + n)
− µ(y)dy = d log S(y) = log
x x S(x)
= log n px .

Edward Furman, Actuarial mathematics MATH3280 – p. 12/66

Taking exponentials:
 Z x+n 
n px = exp − µ(y)dy ,
x
which after the change of variables s = y − x yields
 Z n 
n px = exp − µ(x + s)ds .
0
For a newborn live, i.e. (0)
 Z n 
n p0 = exp − µ(s)ds
0
 Z n 
F (n) = 1 − S(n) = 1 − exp − µ(s)ds .
0

Edward Furman, Actuarial mathematics MATH3280 – p. 13/66

 Second fundamental theorem
Theorem 1 Let f be a continuous real-valued function
defined on [a, b]. Let F be the antiderivative of f , i.e.,
d
f (x) = F (x)
dx
for every x ∈ [a, b]. Then, we have that:
Z b
f (x)dx = F (b) − F (a).
a
Corollary 1 It follows that for every x ∈ [a, b],
Z x
F (x) = f (t)dt + F (a),
a
and thus
Z x 
d d
F (x) = f (t)dt + F (a) = f (x).
dx dx a
Edward Furman, Actuarial mathematics MATH3280 – p. 14/66
 Force of mortality - cont.
Back to the force of mortality, we now have that:
  Z n 
0 d
F (n) = f (n) = 1 − exp − µ(s)ds
dn 0
 Z n 
= exp − µ(s)ds µ(n) = n p0 µ(n).
0
The pdf of T (x) can be found as follows:
 
d d S(x + n)
fT (x) (n) = n qx = 1−
dn dn S(x)
S(x + n)
 0
S (x + n)

= −
S(x) S(x + n)
= n px µ(x + n), n ≥ 0.

Edward Furman, Actuarial mathematics MATH3280 – p. 15/66

Thus Z ∞
n px µ(x + n)dn = 1.
0
To summarize (Each row shows 4 ways to express the
function in the left column.)
F (x) S(x) f (x) µ(x) R
Rx x
− 0 µ(y)dy
F (x) F (x) 1 − S(x) 0 f (y)dy 1 −R e
R∞ x
− 0 µ(y)dy
S(x) 1 − F (x) S(x) x f (y)dy e Rx
f (x) F 0 (x) −S 0 (x) f (x) µ(x)e− 0 µ(y)dy
F 0 (x) −S 0 (x) f (x)
µ(x) 1−F (x) S(x) S(x) µ(x)
At home, make sure you are able to build similar table for
T (x).

Edward Furman, Actuarial mathematics MATH3280 – p. 16/66

Example 3 Show that
Z x+n
lim µ(y)dy = ∞.
n→∞ x

Solution Note that:

lim n px = 0.
n→∞
Thus:
Z x+n
lim µ(y)dy = lim (−log n px ) = ∞,
n→∞ x n→∞

as needed. 
The above example literally says that we have a restriction
on possible force of mortality functions.

Edward Furman, Actuarial mathematics MATH3280 – p. 17/66

Example 4 Let Ac = Ω\A within a sample space and
P (Ac 6= 0). The following expresses an identity in probability
theory:
P (A ∪ B) = P (A) + P (B ∩ Ac ) = P (A) + P (Ac )P (B|Ac ).
Rewrite this identity in actuarial notations for the events:
A = {T (x) ≤ t} and B = {t < T (x) ≤ 1}, 0 < t < 1.
Solution
P (A ∪ B) becomes P (T (x) ≤ 1) = qx
P (A) becomes t qx
P (B|Ac ) is 1−t qx+t .

At last, we obtain:
qx = t qx + t px · 1−t qx+t .

Edward Furman, Actuarial mathematics MATH3280 – p. 18/66

 To conclude:
Rewrite the expression for P (B ∩ Ac )from Example 4 in
actuarial notations.
You have to solve all the exercises on the website and
answer questions on the slides.

References
Bowers, N. L., Hickman, J. C., Nesbitt, C. J., Jones, D. A.
and Gerber, H. U. (1997). Actuarial mathematics, 2nd
edition, Society of Actuaries, Itasca, Illinois.

Edward Furman, Actuarial mathematics MATH3280 – p. 19/66

 Life table functions
Denote by l0 a group of newborns (e.g. l0 = 100000). Each
newborn’s age-at-death has a distribution specified by S(x).
Also, let L(x) be the rv denoting the number of survivors to
age x, generally indexed by j = 1, 2, . . . , l0 . Then:
l0
X
L(x) = 1(j),
j=1

where 1(j) is the indicator function for the survival of life j

that is 1(j) = 1 if this life survives to age x and 0 otherwise.
Definition 3 We denote by lx the expected number of
survivors to age x from the l0 newborns, and we formulate
this as: (E[1(A)] = P (A))  
l0
X
lx = E[L(x)] = E  1(j) = l0 S(x).
j=1
Edward Furman, Actuarial mathematics MATH3280 – p. 20/66
Does the latter expectation remind you an expectation
of a well known rv?
Note that by general reasoning we have that:
lx = l0 S(x)
as well.
Let n D(x) = L(x) − L(x + n) be the rv representing the
group of deaths between ages x and x + n.
Definition 4 We denote by n dx the expected number of
deaths between ages x and x + n among the initial l0 lives:
n dx = E[n D(x)] = l0 (S(x) − S(x + n)) = lx − lx+n .
We can express the force of mortality in terms of lx , l0 :
S 0 (x) dlx
µ(x) = − =− .
S(x) lx dx
Edward Furman, Actuarial mathematics MATH3280 – p. 21/66
We can further easily produce (show at home):
 Z x 
lx = l0 exp − µ(y)dy ,
0
 Z x+n 
lx+n = lx exp − µ(y)dy ,
x
Z x+n
lx − lx+n = ly µ(y)dy.
x
Example 5 Show that the local extreme points of lx µ(x)
correspond to points of inflection of lx .
Proof Indeed,
d d dlx d2
lx µ(x) = − lx = − 2 lx .
dx dx lx dx dx

Edward Furman, Actuarial mathematics MATH3280 – p. 22/66

Figure of lx

Figure 1: Plot of lx

Edward Furman, Actuarial mathematics MATH3280 – p. 23/66

Figure of lxµ(x)

Figure 2: Plot of lx µ(x)

Edward Furman, Actuarial mathematics MATH3280 – p. 24/66

Figure of a force of mortality

Figure 3: Plot of µ(x)

Edward Furman, Actuarial mathematics MATH3280 – p. 25/66

 Fractional ages
Life table functions investigated hitherto specify the cdf of
K(x) completely. To specify the cdf of (T (x) we must
postulate an analytic form or adopt an assumption in
addition to the life table functions we had.
We will further review three different assumptions for
fractional ages:
1. Linear interpolation,
S(x + t) = (1 − t)S(x) + tS(x + 1).
2. Exponential interpolation,
log S(x + t) = (1 − t) log S(x) + t log S(x + 1).
3. Harmonic interpolation,
1/S(x + t) = (1 − t)/S(x) + t/S(x + 1).

Edward Furman, Actuarial mathematics MATH3280 – p. 26/66

Linear interpolation

Figure 4: Linear interpolation for lx+s , 0 < s < 1

Edward Furman, Actuarial mathematics MATH3280 – p. 27/66

Due to the plot, we can find the value of Lx+s from the
following equations:
lx − lx+s x+s−x
= = s,
lx − lx+1 x+1−x
which yield
lx+s = lx − slx + slx+1 .
Finally, we find that:
lx+s = (1 − s)lx + slx+1 .

Edward Furman, Actuarial mathematics MATH3280 – p. 28/66

Linear interpolation:
lx+s = (1 − s)lx + slx+1 = lx − sdx ,
where dx = lx − lx+1 . Further, dividing by lx , we have that
s px = 1 − sqx ⇒ s qx = sqx .
As qx is tabulated we can calculate s qx for any non-integer
duration s.
Also, we have that
Z s Z s
s qx = P (T (x) ≤ s) = fT (x) (t)dt = t px µx+t dt = s · qx
0 0
and thus
d
fT (x) (s) = s qx = qx , for 0 < s < 1.
ds
Edward Furman, Actuarial mathematics MATH3280 – p. 29/66
As fT (x) (s) is constant in s and equal to qx , deaths are said
to be uniformly distributed over the interval (x, x + 1).
We have seen that
f (x) f (x) fT (x) (s)
µ(x) = = and, similarly, µ(x + s) = .
S(x) x p0 s px
Then, the force of mortality is
qx qx
µ(x + s) = = ,
s px 1 − s · qx
which increases in s.
If both the age and the duration are non-integer, i.e., we
want to calculate s−t qx+t , 0 < t < s < 1, then
s px
s px =t px ·s−t px+t ⇒ s−t px+t = .
t px

Edward Furman, Actuarial mathematics MATH3280 – p. 30/66

Hence,
s px 1 −s q x
s−t qx+t = 1 −s−t px+t =1− =1− ,
t px 1 −t q x
which after applying the UDD assumption reduces to
1 − s · qx (s − t)qx
s−t qx+t =1− = , for 0 < t < s < 1.
1 − t · qx 1 − t · qx

Edward Furman, Actuarial mathematics MATH3280 – p. 31/66

Example 6 Given p90 = 0.75, calculate under the UDD
assumption 1 q90 and 1 q90 11 .
12 12 12

Solution We have that

1 1 1
1 q90 = · q90 = · (1 − p90 ) = · 0.25 = 0.02083.
12 12 12 12
Also, taking t = 11
12 and s = 1, we arrive at
11 1
(1 − 12 )q90 12 0.25
s−t qx+t = 1 q90 11 = 11 = = 0.027027.
12 12 1− 12 q90 1 − 11
12 0.25

Edward Furman, Actuarial mathematics MATH3280 – p. 32/66

Example 7 For two lives aged (x) with independent future
lifetimes k| qx = 0.1(k + 1), k = 0, 1, 2 UDD is assumed. What
is the probability that both lives will survive 2.25 years.
Solution Let us calculate the probability that (x) will survive
2.25 years. Let lx = 1, then using dx+k = lx · k| qx , we get
lx+1 = lx − dx = 1 − 0.1 = 0.9;
lx+2 = lx+1 − dx+1 = 0.9 − 0.2 = 0.7;
lx+3 = lx+2 − dx+2 = 0.7 − 0.3 = 0.4.
Linearly interpolating between lx+2 and lx+3 , we obtain that
lx+2.25
lx+2.25 = (1−0.25)0.7+0.25·0.4 = 0.625, ⇒ 2.25 px = = 0.625.
lx
As lives are independent the answer is 0.6252 .

Edward Furman, Actuarial mathematics MATH3280 – p. 33/66

Graphically, the main ideas of UDD can be seen as

Figure 5: lx+s decreases linearly and µx+s increases.

Edward Furman, Actuarial mathematics MATH3280 – p. 34/66

 Constant force of mortality - CFM
The force of mortality can be used to specify the distribution
of X . Indeed,
−S 0 (x) d
µ(x) = = − logS(x)
S(x) dx
Z x+s  
S(x + s)
− µ(y)dy = log = logs px
x S(x)
 Z x+s 
s px = exp − µ(y)dy or for t = y − x
 Zx s 
= exp − µ(x + t)dt .
0
Under the CFM assumption, we find the value of µ as
 Z 1 
px = exp − µ(x + t)dt = exp(−µ) ⇒ µ = −log(px ).
0
Edward Furman, Actuarial mathematics MATH3280 – p. 35/66
Further, for 0 < s < 1, we have that
s px = exp(−sµ) = (px )s .
Also, if both the age and the duration are non-integer, then
for 0 < t < s < 1, we have that
 Z s 
s−t qx+t = 1 −s−t px+t = 1 − exp − µ(x + r)dr
t
 Z s 
= 1 − exp − µdr = 1 − exp(−(s − t)µ).
t
It follows that s−t qx+t =s−t qx . why?
At home show that the CFM is consistent with
exponential interpolation.

Edward Furman, Actuarial mathematics MATH3280 – p. 36/66

Example 8 Given p90 = 0.75, calculate under the CFM
assumption 1 q90 and 1 q90 11 .
12 12 12

Solution First we find the constant µ over the interval

(90,91). Namely,
µ = −log(p90 ) = −log(0.75) = 0.287682.
Then, we have that
1
1 q90 = 1 − exp(− µ) = 0.023688.
12 12
Further, for t = 11 12 and s = 1,
     
11 1
1 q 11 = 1−exp − 1− µ = 1−exp − µ = 0.023688.
90
12 12 12 12

Edward Furman, Actuarial mathematics MATH3280 – p. 37/66

Graphically, the main ideas of CFM can be seen as

Figure 6: lx+s decreases and µx+s is constant.

Edward Furman, Actuarial mathematics MATH3280 – p. 38/66

 The Balducci assumption
Hyperbolic interpolation: for 0 < s < 1
1 1−s s (1 − s)lx+1 + slx
= + = ,
lx+s lx lx+1 lx · lx+1
which implies that
lx · lx+1 lx+1
lx+s = = ,
lx+1 + sdx px + sqx
and dividing by lx , we obtain that
px px
s px = = .
px + sqx 1 − (1 − s)qx
Also,
1 − (1 − s)qx − px sqx
s qx = = .
1 − (1 − s)qx 1 − (1 − s)qx

Edward Furman, Actuarial mathematics MATH3280 – p. 39/66

More generally, for both the age and the duration being
non-integer,

lx+t+s lx+t+s /lx t+s px 1 − (1 − t) qx

s px+t = = = = .
lx+t lx+t /lx t px 1 − (1 − (t + s)) qx
And thus,
1 − (1 − t) qx sqx
s qx+t =1− = .
1 − (1 − (t + s)) qx 1 − (1 − (t + s)) qx
Notice, that the denominator increases in t and hence s qx+t
decreases in age. Plausible?
 
d d px qx
µx+s = − log s px = − log = .
ds ds 1 − (1 − s) qx 1 − (1 − s) qx
The force of mortality decreases over the year.
Edward Furman, Actuarial mathematics MATH3280 – p. 40/66
Example 9 Let qx = 0.1 and assume that mortality follows
the Balducci assumption. Calculate 1 qx+ 1 .
2 4
Solution Straightforward plugging s = 1/2 and t = 1/4 into
the general formula, yelds
(1/2) qx 0.05
1 qx+ 1 = = .
2 4 1 − (1 − (1/2 + 1/4)) qx 0.975

Edward Furman, Actuarial mathematics MATH3280 – p. 41/66

Graphically, the main ideas of the Balducci assumption can
be seen as

Figure 7: lx+s decreases and µx+s decreases.

Edward Furman, Actuarial mathematics MATH3280 – p. 42/66

 To conclude:
You should remember and be able to derive:

Function UDD CFM Balducci

lx+1
lx+s lx − sdx lx ·s px px +sqx
sqx
s qx sqx 1 −s p x 1−(1−s)qx
sqx sqx
s qx+t 1−tqx 1 −s p x 1−(1−(s+t))qx
qx qx
µ(x + s) 1−sqx − log px 1−(1−s)qx

Edward Furman, Actuarial mathematics MATH3280 – p. 43/66

And some plots:

Figure 8: Fractional ages’ concluding plot.

Edward Furman, Actuarial mathematics MATH3280 – p. 44/66

Some analytical laws of mortality
De Moivre (1725) used a one parameter formula:
µ(x) = (ω − x)−1 , S(x) = 1 − x/ω, 0 ≤ x < ω.
Gompertz (1825) used a two parameter formula:
 
x B
µ(x) = Bc , S(x) = exp − (cx − 1) , B > 0, c > 1, x ≥ 0.
log(c)
Makeham (1867) used the following:
 
B
µ(x) = A + bcx , S(x) = exp −Ax − (cx − 1) ,
log(c)
where B > 0, A ≥ −B, c > 1, x ≥ 0.
Weibull (1951) suggested:

x B+1 
B
µ(x) = Ax , S(x) = exp −A , A > 0, B > 0, x ≥ 0.
B+1
Edward Furman, Actuarial mathematics MATH3280 – p. 45/66
 A representation of E[X]
Let X v F and X ∈ R+ , then using integration by parts we
have that:
Z ∞ Z ∞
E[X] = xf (x)dx = − xdF (x)
0
 Z ∞0 
= − xF (x)|∞ 0 − F (x)dx
0
 Z ∞  Z ∞
= − 0− F (x)dx = F (x)dx
0 0
that is the ’tail’ representation of E[X].
Prove the following ’quantile based’ form:
Z 1
−1
E[X] = F (q)dq.
0

Edward Furman, Actuarial mathematics MATH3280 – p. 46/66

 Some characteristics of life tables
◦
Definition 6 The expected value of T (x) is denoted by ex
and is referred to as the complete expectation of life.
Due to the above definition and from the tail representation
of the mathematical expectation:
Z ∞
◦
ex = E[T (x)] = t px dt.
0
It can be easily shown (just as in the case of E[X]) that:
Z ∞
E[X 2 ] = 2 xF (x)dx,
0
thus leading to:
Z ∞
E[T (x)2 ] = 2 tt px dt.
0

Edward Furman, Actuarial mathematics MATH3280 – p. 47/66

For the variance, we have that:
Var[T (x)] = E[T (x)2 ] − E2 [T (x)]
Z ∞  ◦ 2
= 2 tt px dt − ex .
0
Definition 7 The median future lifetime of (x) is denoted by
m(x), and it can be found by solving:
S(x + m(x))
P (T (x) > m(x)) = m(x) px = = 0.5
S(x)
for m(x).
Definition 8 The mode for (x) can be obtained as:
arg max t px µ(x + t).
t∈R+

Edward Furman, Actuarial mathematics MATH3280 – p. 48/66

Definition 9 The curtate expectation of life table is denoted
by ex and is given by:
∞
X ∞
X
E[K(x)] = k k px qx+k = k px .
k=0 k=1
To prove the latter identity, first define the forward difference
operator by:
∆f (x) = f (x + 1) − f (x),
and second recall the summation by parts formula:
Xn X n
f (k)∆g(k) = [f (k)g(k)] |n+1
m − g(k + 1)∆f (k).
k=m k=m

Edward Furman, Actuarial mathematics MATH3280 – p. 49/66

Then:
∞
X
ex = kP (K(x) = k)
k=0
X∞
= k (P (K(x) ≤ k) − P (K(x) ≤ k − 1))
k=0
X∞
= k (k+1 qx − k qx )
k=0
X∞ ∞
X
= k∆k qx = k∆(1 − k px )
k=0 k=0
X∞
= − k∆k px .
k=0
Edward Furman, Actuarial mathematics MATH3280 – p. 50/66
And hence, using the summation by parts:
∞
!
X
ex = − k k px |∞
0 − k+1 px ∆k
k=0
∞
X ∞
X
= k+1 px = k px ,
k=0 k=1
bearing in mind that ex is finite. 
Applying the same technique, we arrive at:
∞
X
E[K(x)2 ] = k 2 k px qx+k ,
k=0
thus yielding

Edward Furman, Actuarial mathematics MATH3280 – p. 51/66

 ∞
X
= − k 2 ∆(k px )
k=0
∞
X
= −k 2 k px |∞
0 + p
k+1 x ∆k 2

k=0
∞
X
= (2k + 1)k+1 px
k=0
X∞
= (2k − 1)k px .
k=1
Finally, in this case:
∞
X
Var[K(x)] = (2k − 1)k px − e2x .
k=1
Edward Furman, Actuarial mathematics MATH3280 – p. 52/66
 More characteristics
Definition 10 The function Tx is defined as:
Z ∞ Z ∞
Tx = lx+t dt = ly dy,
0 x
and it represents the expected total future lifetime of a
group of lx individuals all aged x.
To understand the interpretation, recall that:
Z ∞ Z ∞
◦ lx+t
ex = t px dt = dt.
0 0 lx
Thus, due to Definition 10:
◦
Tx = lx ex ,
motivating the definition.

Edward Furman, Actuarial mathematics MATH3280 – p. 53/66

Definition 11 The function Lx is defined as:
Z 1 Z x+1
Lx = lx+t dt = ly dy,
0 x
and it denotes the total expected time lived between ages x
and x + 1 by lx lives aged x.
This time the interpretation can be seen from, e.g.,
Lx = Tx − Tx+1 .
Definition 12 The central rate of mortality is defined as:
R1
0 lx+t µ(x + t)dt dx
mx = R1 = ,
Lx
0 lx+t dt
which is the weighted average force of mortality
experienced between ages x and x + 1 with weight lx+t at
age x + t.
Edward Furman, Actuarial mathematics MATH3280 – p. 54/66
 Recursion formulas
◦
Example 10 Prove the following recursions for ex , ex :
c(x) 1
u(x) = c(x) + d(x)u(x + 1) and u(x + 1) = − + u(x).
d(x) d(x)
Solution Let us first check the left hand side for ex :
∞
X X∞ ∞
X
ex = k px = px + k px = px + px k px+1 = px + px ex+1 ,
k=1 k=2 k=1
◦
thus u(x) = ex , c(x) = d(x) = px , u(∞) = 0. In the case of ex :
Z ∞ Z 1 Z ∞
◦
ex = t px dt = t px dt + t px dt
0 0 1
Z 1 Z ∞ Z 1
◦
= t px dt + px t px+1 dt = t px dt + px ex+1 .
0 0 0

Edward Furman, Actuarial mathematics MATH3280 – p. 55/66

 Elementary numerical integration

Figure 9: Trapezoidal approximation for integrals.

According to the plot, the trapezoidal approximation is:

Z b
1
f (x)dx ≈ (b − a)f (b) + (b − a) (f (a) − f (b))
a 2
1
= (b − a) (f (a) + f (b)) .
2
Edward Furman, Actuarial mathematics MATH3280 – p. 56/66
 Recursion formulas - cont.
Thus, we have that
Z 1
◦
u(x) = ex , c(x) = t px dt, d(x) = px , u(∞) = 0.
0
Here, we can approximate c(x) with the trapezoidal
approximation:
Z 1
1 1
c(x) = t px dt ≈ px + (1 − px ) = (1 + px ) .
0 2 2
How do we obtain the second approximation?

Edward Furman, Actuarial mathematics MATH3280 – p. 57/66

 One more example
Example 11 Show that under the UDD assumption, we
have that:
◦
ex = ex + 0.5.
Solution Due to Definition 6, we have that:
Z ∞ ∞ Z 1
◦ 1 1 X
ex = lx+t dt = lx+k+t dt
lx 0 lx
k=0 0
∞ Z 1
1 X
≈ ((1 − t)lx+k + tlx+k+1 ) dt
lx
k=0 0
∞  
1 X 1 1
= lx+k + lx+k+1
lx 2 2
k=0
∞
!
1 1 X 1
= lx + lx+k = + ex 
lx 2 2
k=1
Edward Furman, Actuarial mathematics MATH3280 – p. 58/66
 Select life tables
Assumption 2 Let (x) possess a different probability than
the general population of surviving to the next year for each
of the first k years after selection (the select period).
Definition 13 We denote by:
t p[x]+r = 1 − t q[x]+r
the probability that (x + r) selected at age x will survive at
least next t years.
All the relations we have seen hitherto hold, i.g.
t+s p[x]+r = t p[x]+r · s p[x]+r+t .
Assumption 3 Persons aged (x), selected before a number
of years that is bigger than r have the same survival
probabilities, i.e., for all j > 0,
t p[x]+r = t p[x−j]+r+j .
Edward Furman, Actuarial mathematics MATH3280 – p. 59/66
In simpler words, if k = 2, then:
q[x]+2 = q[x−1]+3 = q[x−2]+4 = · · · = qx+2 .
In general, the select life table is constructed by working
back from lx+r successively to l[x]+r−1 , l[x]+r−2 , . . . , l[x] using
the select probabilities p[x]+r−1 , p[x]+r−2 , . . . , p[x] . More
precisely, for all r ≥ 0,
l[x]+r+1
l[x]+r = .
p[x]+r
Example 12 Let k = 1 and an individual is aged x on
selection. Then l[x] = lx+1 /p[x] . Also, l[x]+1 = lx+1 for k = 1.
We can further find e.g.
∞
◦ 1 1 1 X 1 lx
e[x] ≈ + e[x] = + lx+r = + ex ,
2 2 l[x] 2 l[x]
r=1 Edward Furman, Actuarial mathematics MATH3280 – p. 60/66
which after another approximation reduces to:
 
◦ 1 lx ◦ 1
e[x] ≈ + ex − .
2 l[x] 2
Example 13 Let k = 2 and an individual is aged x on
selection. Then l[x]+1 = lx+2 /p[x]+1 and l[x] = l[x]+1 /p[x] . Also,
∞
!
◦ 1 1 1 X
e[x] ≈ + e[x] = + l[x]+1 + lx+r
2 2 l[x]
r=2
1 1

= + l[x]+1 + lx+1 ex+1
2 l[x]
  
1 1 ◦ 1
≈ + l[x]+1 + lx+1 ex+1 − .
2 l[x] 2

Edward Furman, Actuarial mathematics MATH3280 – p. 61/66

Figure 10: A 1967-70 mortality life table for assured lives.
Edward Furman, Actuarial mathematics MATH3280 – p. 62/66
 The deterministic approach
A deterministic survivorship group as it is represented by
the life tables
initially consists of l0 lives aged 0,
is a subject to the annual rate of mortality (decrement) qx ,
is closed, i.e., no entrances are allowed - only
decrements.
Thus the progress of the group is determined by:
l1 = l0 (1 − q0 ) = l0 − d0
l2 = l1 (1 − q1 ) = l1 − d1 = l0 − (d0 + d1 )
..
.
x−1
X
lx = lx−1 (1 − qx−1 ) = lx−1 − dx−1 = l0 − dj .
j=0
Edward Furman, Actuarial mathematics MATH3280 – p. 63/66
The latter expression is further rewritten as:
Px−1 !
j=0 dj
lx = l0 1 − = l0 (1 − x q0 ).
l0
We can formulate this as:
l1 = l0 · p0
l2 = l1 · p1 = l0 · p0 · p1
..
.
x−1
Y
lx = lx−1 · px−1 = l0 pj = l0 · x p0 .
j=0

Edward Furman, Actuarial mathematics MATH3280 – p. 64/66

 To conclude:
Some relations between interest rate functions and the
present topic:
Compound interest Survivorship group
A(t) lx
A(t+n)−A(t) lx −lx+n
n it = A(t) n qx = lx
A(t+∆t)−A(t) lx −lx+∆x
δt = lim∆t→0 A(t)∆t µ(x) = lim∆t→0 lx ∆x
dA(t)
= A(t)dt = − ldl x
x dx

Edward Furman, Actuarial mathematics MATH3280 – p. 65/66

 References
References
Bowers, N. L., Hickman, J. C., Nesbitt, C. J., Jones, D. A.
and Gerber, H. U. (1997). Actuarial mathematics, 2nd
edition, Society of Actuaries, Itasca, Illinois.
De Moivre, A. (1725). Annuities on Lives, 3-125.
Gompertz, B. (1825). “On the nature of the function
expressive of the law of human mortality, and on a new
mode of determining the value of life contingencies”,
Philosophical Transactions of Royal Society, (Series A) 115,
513 - 583.
Makeham, W.M. (1867). “On the law of mortality”, Journal
of the Institute of Actuaries 8, 301-310.
Weibull, W. (1951). “A statistical distribution function of wide
applicability”, Journal of Applied Mechanics 18, 293-297.

Edward Furman, Actuarial mathematics MATH3280 – p. 66/66