1

Week 3

Demography
Outline:
Survival models
Hazard Rates and Survival Models
Actuarial Notation
Life table

Reading
Sherris, Ch 3 (req)
 2

Survival Models
Probability model to calculate
– the probability a life will survive
– expected payments for insurance and annuity contracts
Survival Function:
– Continuous random variable age-at-death X
– Distribution function FX (x)
– Probability density function FX0 (x) = fX (x)
– Survival Function (for a Newborn):
s(x) = Pr (X > x) = 1 FX (x) x 0
– For a life aged 0: FX (0) = 0 so that s(0) = 1
 3

Survival Function for Life Aged x

Note: Conditional probability
– Consider firstly joint probability
Pr (A and B) = Pr (A) Pr (BjA)
= Pr (B) Pr (AjB)
where Pr (AjB) is the probability that A occurs given that B has already occurred.
– Hence the conditional probability
Pr (A and B)
Pr (BjA) =
Pr (A)
Life aged x the probability that the life will survive to age z (z > x) !Conditional
probability
Pr (X > z and X > x) s(z)
Pr (X > zjX > x) = =
Pr (X > x) s(x)
 4

Discussion
x
If s (x) = 1 100 for 0 x < 100
(a) Determine the probability that a life aged 20 will survive to age 65
(b) (*) Find the probability that a life aged 20 will die between 40 and 60.
 5

Future Lifetime
A life aged x is denoted by the symbol (x)
Future lifetime random variable is T (x) = X x
International Actuarial Notation
qx = Pr [(x) will die within a year] = Pr [T (x) 1]

px = Pr [(x) will survive at least a year] = Pr [T (x) > 1]

t qx = Pr [(x) will die within a t years] = Pr [T (x) t] t 0

t px =1 t qx = Pr [T (x) > t]
 6

Survival Functions and Actuarial Notation

Notation:

t p0 = s(t)

x+t p0 s(x + t)
t px = =
x p0 s(x)

s(x + t)
t qx = 1
s(x)
tju qx
represents the probability of a Life aged x; surviving t years and dying within the
following u years:
 7

– Actuarial notation tjuqx

tju qx = Pr [t < T (x) t + u]

s(x + t) s(x + t + u)
=
s (x)
s(x + t + u) s(x + t)
= 1 1
s (x) s (x)
= t+uqx t qx
 8

Hazard Rate
Survival models - key feature is hazard rate or failure rate function
In actuarial science and demography the hazard or failure rate is referred to as the
force of mortality
Hazard rate definition
d
dx s (x)
(x) =
s (x)
s0 (x)
=
s (x)
fX (x)
=
1 FX (x)
Actuarial notation - force of mortality is denoted by x
 9

By definition
d
dy s (y) d
(y) = = ln s (y)
s (y) dy
Integrating from x to x + t we have
Z x+t Z x+t
s (x + t)
(y) dy = d ln s (y) = ln
x x s (x)
Z x+t
s (x + t)
= exp (y) dy
s (x) x
= t px
and
Z x
s(x) = exp (y) dy
0
 10

Life Table
The life table is a table showing for each age x:
qx the probability of death between age x and x + 1;
dx the (expected) number of deaths aged x last birthday,
lx the (expected) number alive at exact age x,
Lx the (expected) number living aged between x and x + 1 (x last birthday), also the
(expected) number of years lived by the lx lives aged x over the year from age x to
x + 1,
Tx the number of lives aged x or greater, also the (expected) total future lifetime of
the lx lives aged x;
ex the average future lifetime of a life aged x, also referred to as the complete expec-
tation of life.
 11

Lx is a measure of the number of lives exposed-to-risk of dying aged x last birthday

Any life in this group who dies will be classified as a death at age x:

Z 1
lx + lx+1
Lx = lx+tdt
0 2
Note, by definition
1
X
Tx = Lx+t
t=0

Tx is the total number of future years lived by the lx lives aged x so that the average
number of years lived is Tlxx
 12

Excel - Examples of Life Tables

IA64-70 Table for Australian Insured Lives
– based on the mortality of whole-of-life policies for mainly male lives from data con-
tributed by 14 Life Offices
– The data :January 1964 to December 1970
– shows "ultimate" mortality rates - lives with policies for more than 2 years
Australian males A90-92 males
– Population mortality table: male lives in Australia in the 1991 census,
– and male deaths over the period 1990 to 1992.
– These rates are smoothed using an actuarial technique called graduation (covered
in Actuarial Statistics subject)
 13

Construction
Insurance company and population census and deaths data provides the information
to calculate the qx rates
Table starts with a radix which is the base for the table eg For (A90-92 males) the
radix is l0 = 1; 000; 000.
Recursive relationship:
dx = qxlx

lx+1 = lx dx

lx + lx+1
Lx =
2
Tx = Tx+1 + Lx
 14

lx values are interpreted as the (expected) number of lives exact age x surviving from
the initial l0 lives
Consider a group of l0 new borns then the probability that any one of these will
survive to age x is s(x)
Expected number of lives out of the initial l0 lives who survive to exact age x will be
given by
lx = l0s(x)
The variance of the number of lives surviving to age x will be
l0s(x) [1 s (x)]
Note also that
lx+t
t px =
lx
 15

Example
Use the IA64-70 Life Table to calculate:
(a) the expected age at death of a life aged 20
(b) the probability that a life age 20 will survive to age 40
(c) the probability that a life aged 20 will die within the next 10 years.
 16

Laws of Mortality - De Moivre

Simple models have been proposed for the hazard rate for human lives - referred to
as laws of mortality
De Moivre’s Law - in 1725 Abraham de Moivre assumed the number alive according
to the life table decreased in arithmetical progression
x
s (x) = 1 0 x<!
!
Hazard rate (force of mortality) given by
s0 (x) 1
(x) = = 0 x<!
s (x) ! x
 17

Gompertz
Benjamin Gompertz in 1825 hypothesized that the force of mortality increases with
age in geometrical progression
Hazard rate
(x) = Bcx B > 0; c > 1, x 0
Chance of death increases more and more rapidly with age
Z x
s (x) = exp (y) dy
Z0 x
= exp Bcy dy
0
x
B y
= exp c
ln c 0
B x
= exp (c 1)
ln c
 18

Fitting Laws of Mortality to Life Table Data

Parameters of the Gompertz curve are B and c
Least squares selects the numerical values of the parameters by minimizing the sum
of the squared differences between the Law values and the actual values
Denote actual values for s(x) from the life table by sa (x)
Over the age range xl to xu select values of B and c that minimize the function
xu
X 2
B x
sa (x) exp (c 1)
xl
ln c

Use a spreadsheet such as Excel and the Solver

 19

Fitting Gompertz
Australian IA64-70 Life Table over the ages 10 to 110
Gompertz curve using least squares was given by
(x) = 0:000046 (1:100837)x
Up to around age 80 the Gompertz curve provides a reasonable fit to the IA64-70
data
 20

Makeham
1867 - Makeham suggested an addition of a constant to account for accidents and
infections as well as an increase in hazard geometrically with age
Hazard function is
(x) = A + Bcx
A > B , B > 0;
c > 1, x 0
IA64-70 Life Table over the ages 10 to 110
Best fit Makeham curve was
(x) = 0:0003225 + 0:000031 (1:1066413)x