Contents
1 Future lifetime: probabilities and mortality
2 Actuarial notation
3 Force of mortality
4 Expectation of life
5 Life tables
6 Initial and central rates of mortality
7 Assumptions for fractions of year
8 Comparisons between fractional year assumptions
9 Select tables
10
Future lifetime: probabilities and mortality
Introduction
Lifetime: time to the occurrence of a certain event, e.g.
Lifetime in the context of human mortality (time to death)
the length of the time that a surviving individual will hold an insurance policy (time to withdrawal)
the length to time an individual remains healthy (time to the occurrence of sickness)
Survival models are models that (hopefully) describe the probability distribution of such lifetimes. We review here the related
main concepts and their notations.
Reading: ACTED CT4 Chapter 7
A simple model of survival (continuous)
Future lifetime T of a newborn:
T is a continuous random variable on the interval [0, ], where is the limiting age
(highest age recorded 122 years and 164 days - Jeanne Calment - see, e.g. this link).
The distribution function of T is F (t) = Pr[T t].
The survival function of T is S(t) = Pr[T > t] = 1 F (t).
The probability density function of T is f (t) =
dF (t)
dt .
Future lifetime Tx after age x:
(0 x )
This is for a life who has already survived to age x.
Tx is a continuous random variable taking values in [0, x]
The distribution function of Tx is Fx (t) = Pr[Tx t].
The survival function of Tx is Sx (t) = Pr[Tx > t] = 1 Fx (t).
1
The probability density function of Tx is fx (t) =
dFx (t)
dt .
S(x) S(x + t)
S(x)
S(x + t)
Sx (t) = Pr[Tx > t] = Pr[T > x + t|T > x] =
S(x)
Fx (t) = Pr[Tx t] = Pr[T x + t|T > x] =
Remember:
Pr[A B] = Pr[A] Pr[B|A]
Pr[B|A] =
Pr[A, B]
Pr[A]
Actuarial notation
Actuarial notation
Probability of death
Z
t qx = Pr[Tx t] = Fx (t) =
fx (s)ds
0
Probability of survival
t px
= Pr[Tx > t] = 1 t qx = Sx (t)
Deferred probability
n|m qx
Customary notation: qx = 1 qx , px = 1 px ,
|m qx
= P r[n < Tx < n + m]
= 1|m qx
Note:
s+t px
t px
Exercises
Assume a survival function S(x) = 1
= (t px )( s px+t )
= px px+1 px+t1
x
120 .
1. Find:
(a) p50
(b) 2 p50
(c)
10 q75
d
2. Show that fx (t) = dt
t px .
3. Consider the following probabilities:
p20 = 0.95, 2 p20 = .9, 2 p22 = 0.8.
Find 3 p21 .
2
Force of mortality
Force of mortality
The force of mortality at age x (0 x < )
dF (x)
x = lim+
h0
1
Pr[T x + h|T > x] = dx
h
S(x)
is an instantaneous measure of mortality at age x. It is known as hazard rate in statistics. Looking forward,
x+t
=
=
1
Pr[T x + t + h|T > x + t]
h
1
lim
Pr[Tx t + h|Tx > t].
+
h0 h
lim
h0+
Cumulative Hazard Function
Z
t dt = log S(x),
H(x) =
0
so that
Z
S(x) = exp
t dt
and
t px
Z
= exp
x+t
s ds .
Examples
Suppose x = 0.02 for 3 x 10.
1. Calculate the probability that a life aged exactly 5 will die before age 10.
2. Calculate the probability that a life aged exactly 3 will die between the ages of 8 and 10.
3
An important result:
fx (t) = t px x+t
Proof:
fx (t)
=
=
=
=
=
d
Pr[Tx t]
dt
lim
1
(Pr[Tx t + h] Pr[Tx t])
h
lim
1
(Pr[T x + t + h|T > x] Pr[T x + t|T > x])
h
h0+
h0+
lim
Pr[T x+t+h]Pr[T x]
S(x)
lim
Pr[T x+t]Pr[T x]
S(x)
h0+
h0+
Pr[T x + t + h] Pr[T x + t]
S(x)h
S(x + t)
1 Pr[T x + t + h] Pr[T x + t]
lim+
S(x)
h
S(x + t)
h0
= Sx (t)x+t
=
t px x+t
Expectation of life
The complete expectation of life ex at age x is defined by
ex
E[Tx ]
Z
tfx (t)dt
0
t px )dt
t
Z x
x
= [tt px ]t=0 +
t px dt
=
t(
Z
=
x
t px dt
The curtate future lifetime of a life age x is defined by
Kx = [Tx ] (the integer part of Tx )
.
Kx is a discrete random variable taking integer values 0, 1, , [ x].
Pr[Kx = k] = Pr[k Tx < k + 1] = k px qx+k
The curtate expectation of life ex is defined by
[x]
ex
E[Kx ] =
k k px qx+k
k=0
1 px qx+1
2 px qx+2
+ 2 px qx+2
3 px qx+3
+ 3 px qx+3
+ 3 px qx+3
+
+
[x] px qx+[x]
+ [x] px qx+[x] + + [x] px qx+[x]
[x] [x]
[x] [x]
X X
X X
k=1
j px qx+j
j=k
k=1
j=k
[x]
j| qx
k px ,
k=1
We has used the following result:
[x]
[x]
j| qx
j=k
Pr(j Tx < j + 1)
j=k
[x]
Pr(j=k {j Tx < j + 1})
Pr({k Tx < [ x] + 1})
Pr(k Tx )
k px .
Exercises
1. Show ex = px (1 + ex+1 )
2. Suppose e50 = 30 and 50+t = 0.05 for 0 t 1. What is e51 ?
5
Life tables
The Life Table
Radix of the table l0 : a hypothetical cohort group of new-born lives
lx = l0 S(x): the expected number of survivors to age x out of the original group of l0 new-borns
The Life Table
n px
lx+n
lx
conditional probability of surviving to age x + n, given alive at age x
dx = lx lx+1 the expected number who die between ages x and x + 1
n dx
= lx lx+n the expected number who die between ages x and x + n
n qx
x =
n dx
lx
: conditional probability of dying within n years, given alive at age x
d
S(x)
dx
S(x)
d
dx
lx
lx
Numerical examples
x
lx
qx
60 86714 0.01392
61 85507 0.01560
62 84173 0.01749
63 82701 0.01965
64 81076 0.02199
ex
17.850
17.095
16.357
15.640
14.943
1. Use the above Life Table to find:
(a) p60
(b) 2 p60
(c) 4 q60
2. True or False? Suppose that at time 0 there are 86714 lives aged 60 (exactly). At time 4 there will be 81076 lives still
alive.
6
The Life Table: deferred quantities
Generally
n|m qx
=
=
=
=
P r[n < Tx < n + m]
lx+n lx+n+m
lx
lx+n lx+n lx+n+m
lx
lx+n
n px m qx+n
so that
n| qx
= n|1 qx = n px qx+n
and
n| qx
= P r[Kx = n] = P r[n Tx < n + 1] =
dx+n
lx
Numerical Example
Use the Life Table ELT
15 (Female) (above) to find
5|5 q60 .
Initial and central rates of mortality
Initial and central rates of mortality
Initial rate of mortality
probability that a life alive at age x (the initial time) dies before age x + 1
denoted qx
We have
qx =
The central rate of mortality
dx
.
lx
probability that the population dies before age x + 1
denoted by mx
useful in population projections
We have
mx
=
=
=
dx
R x+1
lz dz
x
qx
,
R1
p dt
0 t x
P r[Tx 1]
R1
p dt
0 t x
R1
p dt
0 t x x+t
R1
p dt
0 t x
Discussion
Discuss the relationship between the central and initial rates of mortality: which one is greater?
Assumptions for fractions of year
Methods for non-integral ages
All these functions s px can be determined from a life table for integral x and s (using the formula s px =
non-integral s.
lx+s
lx ),
but not for
The determination of functions for non-integral ages requires that values of lx+s be available for all s, 0 s 1 and
integral x.
To this end, we will assume that lx+s (0 s 1, x is any integer) has one of the following mathematical form:
Linear form for lx+s (Uniform distribution of deaths)
Exponential form for lx+s (Constant force of mortality)
Hyperbolic form for lx+s (Balducci assumption/distribution)
We also assume that lx+s is differentiable on the open interval 0 < s < 1, which allows us to evaluate x+s and hence
the conditional density function f (s|X > x) = s px x+s .
Uniform distribution of deaths (UDD) (linear form for lx+s )
We have
lx+s
= lx s dx = s lx+1 + (1 s) lx
(linear interpolation)
lx+s
lx s d x
=
= 1 s qx
=
s px
lx
lx
f (s|X > x) = qx
s qx
ex
t px+s
1 s px = s qx
1
= ex +
2
=
lx (s + t) dx
lx+s+t
=
for 0 s, t 1 and s + t 1
lx+s
lx s dx
Example
Assume a UDD. Given that p62 = 0.99219, 2 p63 = 0.98171, and p65 = 0.98926, calculate 3.5 p62.3 .
We have
q62 = 1 p62 = 0.00781 q65 = 1 p65 = 0.01074
1 q62
= 0.99452
0.7 p62.3 =
1 0.3q62
1 0.8q65
= 0.99141
0.8 p65 =
1 0q65
3.5 p62.3
(0.7 p62.3 )(2 p63 )(0.8 p65 )
0.99452 0.98171 0.99141 = 0.96794
Constant force of mortality (exponential form for lx+s )
For any integer x and 0 s 1, assume that
lx+s is a continuous exponential function with respect to s on s [0, 1], i.e. of form a bs
in other words we assume that the log of lx is linear:
ln lx+s = ln a + ln b s.
We have then
log lx+s
(1 s) log lx + s log lx+1
x+s
x for 0 < s < 1
s px
esx
ts px+s
e(ts)x
Balducci assumption (hyperbolic form for lx+s )
For any integer x and s [0, 1], assume that
lx+s is a hyperbolic function with respect to s on s [0, 1], i.e. of form (a + bs)1
alternatively,
the function)
1
lx+s
= (1 s)
1
lx
+s
1
lx+1
for 0 s 1 (harmonic interpolation: linear interpolation on the reciprocal of
Then for any integer x and s [0, 1],
a convenient relationship:
Example: Calculate
0.2|0.4 qx
1s qx+s
= (1 s)qx , 0 s 1
if qx = 0.4.
1
0.2|0.4 qx
=
=
=
=
=
lx+0.2 lx+0.6
=
lx
lx lx+1
0.8lx+1 +0.2lx
10.2
0.2
lx + lx+1
1
10.6
0.6
lx + lx+1
lx
lx lx+1
0.4lx+1 +0.6lx
lx
lx+1 /lx
lx+1 /lx
0.8lx+1 /lx + 0.2 0.4lx+1 /lx + 0.6
px
px
0.8px + 0.2 0.4px + 0.6
0.6
0.6
0.8 0.6 + 0.2 0.4 0.6 + 0.6
0.1681.
Comparisons between fractional year assumptions
Comparisons
The idea of UDD is to join lx and lx+1 linearly. What happens with the other assumptions?
The idea of a constant force of mortality is self-explanatory. What happens with the other assumptions?
5000
6000
7000
l_{x+s}
8000
9000
10000
Effect on lx+s , 0 < s < 1
0.0
0.2
0.4
0.6
0.8
1.0
Force of mortality under constant force of mortality
x+s = ln px ,
0<s<1
qx
,
1 sqx
0<s<1
Force of mortality under UDD
x+s =
Force of mortality under Balducci
x+s =
qx
,
1 (1 s)qx
10
0<s<1
0.3
0.4
0.5
mu_{x+s}
0.6
0.7
Effect on x+s , 0 < s < 1
0.0
0.2
0.4
0.6
0.8
1.0
Select tables
Select life tables
Assume now mortality also depends of the date of entry in a group.
[x] + t still means age is (x + t), but
the age at date of joining population is x; and
the duration from the date of joining the population is t
l[x]+t : expected number of lives alive at duration t having joined the population at age [x] based on some assumed radix
s: the length of the select period
After the select period the lives experience ultimate mortality, i.e. l[x]+t = lx+t for t s.
Example A 2-year select-and-ultimate mortality table:
[x]
30
31
32
l[x]
9907
9903
9899
l[x]+1
9905
9901
9896
lx+2
9901
9897
9893
x+2
32
33
34
Examples
Express the following conditional probabilities in terms of ls, assuming a 4-year select period.
1.
2|4 q[20]+1
11
2.
2|4 q[22]+3
You are given the following extract from a 2-year select-and-ultimate mortality table:
[x]
30
31
32
2 p[30]
1| q[31]
2 q[31]+1
l[x]
9907
9903
9899
l[x]+1
9905
9901
9896
lx+2
9901
9897
9893
x+2
32
33
34
l[30]+2
l32
9901
=
=
= 0.9994
l[30]
l[30]
9907
l[31]+1 l33
l[31]+1 l[31]+2
9901 9897
=
=
= 0.0004
l[31]
l[31]
9903
l[31]+1 l[31]+3
l[31]+1 l34
9901 9893
=
=
= 0.0008
l[31]+1
l[31]+1
9901
12
