Universidad Carlos III de Madrid
Licenciatura en Ciencias Actuariales y Financieras
Survival Models and Basic Life Contingencies
PART I. Lecture 3:
Life Tables
1 Introduction
A Life Table is a convenient way of summarising information about a survival
model.
We can define, in very general terms, a Life Table as a study of the mortality
in a sample of persons belonging to a certain population.
The Life Table will lead us to an estimation of the probability distribution
of the random variable: curtate future lifetime of a life aged (x).
2 Basic Functions: lx, dx, px and qx.
The most important function in a Life Table is lx . It is usually the first
column.
lx ≡ Number of persons alive at age x
l0 ≡ Radix or Sample size
It is clear that lx is a decreasing function.
A second important function in a Life Table is
dx ≡ Number of persons dying at age x
dx = lx − lx+1 ⇒ lx = lx+1 + dx
1
With the former functions, the probability function of the curtate future
lifetime of a life aged (x) is easily obtained,
© ª dx+i
P θdx = i = i = 0, 1, 2, ...
lx
also the Survival Function,
lx+i+1
Sθdx (i) = = i+1 px i = 0, 1, 2, ..
lx
lx+j
j px = j = 1, 2, 3, ...
lx
lx+1
1 px =
lx
and, finally, the Cumulative Function,
lx − lx+i+1
Fθdx (i) = 1 − Sθdx (i) =
lx
lx − lx+j
j qx =
lx
dx
1 qx =
lx
However, Life Tables only include,
1 px ≡ px
1 qx ≡ qx
and we can obtain the actuarial functions in an alternative fashion,
j−1
Y
j px = px · px+1 · px+2 · px+3 · · · · · px+j−1 = px+i
i=0
j−1
Y
j qx = 1− px+i
i=0
= qx + px · qx+1 + 2 px · qx+2 + · · · + j−1 px · qx+j−1
j−1
X
= i px · qx+i
i=0
2
Remark 1 What does the following expression mean?
X
ω−x
lx = dx+i
i=0
Remark 2 Notice these two expressions,
j−1
X
j qx = i px · qx+i (Discrete)
i=0
Z j Z j
j qx = fx (z)dz = z px µx+z dz (Continuous)
0 0
3 The study of non-integer instants of death.
The Life Table is a very convenient way of studying the mortality but it is
only designed for estimating the probability distribution of the curtate future
lifetime of life aged (x).
The next question is indeed: What about the continuous considera-
tion?
Let us start studying now the following probabilities,
© ª
P θdx ≤ i = P {θx ≤ i + 1}
and subsequently,
Z i+1
i+1 qx = Fθdx (i) = fθx (z)dz = Fθx (i + 1) = Fx (i + 1)
0
j qx = Fx (j) j = 1, 2, ... 0 qx = Fx (0) = 0
j px = Sx (j) j = 1, 2, ... 0 px = Sx (0) = 1
It is then clear that we will be able to obtain Fx (j) or Sx (j) for integer values
of j and x using the former expression; and what about Sx (j+s), s ∈ (0, 1)?
Firstly, let us use this expression
Sx (j + s) = Sx (j) · Sx+j (s)
3
this means that we will be only interested in the functions
Sk (s),s ∈ (0, 1)
for any integer value of k, because we already know the value of
Sx (j) = j px
Notice that
Sk (s) ∈ (1, Sk (1))
Yes, but How can we get Sk (s), s ∈ (0, 1) ?
We need to use interpolation techniques.
3.1 Linear Interpolation: Uniform Distribution of Deaths
The first interpolation model will be linear
Sk (s) = a + bs
We already know that
½
1 s=0
Sk (s) =
pk s=1
and then,
1 = a
pk = a + b
solving the linear system of equations
Sk (s) ' 1 − s · qk
or Fk (s) ' s · qk
Let us now multiply by lk both sides of the former expression,
lk · Fk (s) = s · lk · qk = s · dk
4
and abusing of the notation,
dk+s = s · dk
which means that deaths are occurring at a constant rate dk with time; in
other words, deaths are uniformly distributed during the year. For instance,
with age k + (1/2), half of the people dying with age k will be dead.
As a consequence,
dFk (s)
fk (s) = = qk
ds
fk (s) qk
µk+s = = (= µk (s))
Sk (s) 1 − sqk
It is important to mention that
qk
lim− µk+s =
s→1 1 − qk
lim+ µk+1+s = qk+1
s→0
This means that a sensible approximation for µk , k integer
qk
1−qk
+ qk+1 2 qk
µk+1 ' =
2 2 · pk
And, as an example
Z ω−x X Z k+1
ω−x−1
o
ex = Sx (z)dz = Sx (z)dz
0 k=0 k
X Z 1
ω−x−1 X Z 1
ω−x−1
= Sx (k + s)ds = Sx (k)Sx+k (s)ds
k=0 0 k=0 0
X
ω−x−1 Z 1
= Sx (k) (1 − s qx+k ) ds
k=0 0
X
ω−x−1 ³ qx+k ´
= p
k x 1 −
k=0
2
5
3.2 Exponential interpolation.
Again, we already know that
½
1 s=0
Sk (s) =
pk s=1
but this time we will adjust a function like this
Sk (s) = a · cs
The system of equations
1 = a
pk = a · c
leads us to
Sk (s) ' (pk )s = e−(− log pk )·s
Fk (s) ' 1 − (pk )s = 1 − e−(− log pk )·s
fk (s) ' (− log pk ) · e−(− log(pk ))·s
fk (s)
µk+s = = − log pk (= µk (s))
Sk (s)
The hazard rate is constant inside the intervals (k, k + 1). It is important to
mention that
lim µk+s = − log pk
s→1−
lim µk+1+s = − log pk+1
s→0+
This means that a sensible approximation for µk , k integer
(− log pk ) + (− log pk+1 ) − log2 pk
µk+1 ' =
2 2
And, again
o X Z 1
ω−x−1
ex = Sx (k)Sx+k (s)ds
k=0 0
X
ω−x−1 Z 1
= Sx (k) e−(− log px+k )·s ds
k=0 0
X
ω−x−1 µ ¶
1 − e−(− log px+k )
= k px
k=0
− log px+k
6
4 Miscellaneous Functions
Let us define now,
Z n Z n
o
n Lx = lx ex:ne = lx Sx (z)dz = ”lx+z ”dz
0 0
as the average number of years lived by the lx lives within n years.
The asymptotic case
Z ω−x
o
lim n Lx = Tx = lx ex = ”lx+z ”dz
n→ω−x 0
can then be defined as the average number of years lived by the lx lives
until the extinction of the collective.
Let us present now an alternative formula for the central-death rate,
lx Fx (n) dx+n
n mθx = o =
lx ex:ne n Lx
