Lec2 PDF

Life tables with one source of decrement
Fractional ages
Life table
Definition 1.1 (Life table.)

We shall call the distribution of (u) its life table.
Example 1.1
◦
[x, x + t) lx qx dx Lx Tx ex
[0, 1) 100, 000 0.01260 1, 260 98, 973 7, 387, 758 73.88
[1, 2) 98, 740 0.00093 92 98, 694 7, 288, 785 73.82
[2, 3) 98, 648 0.00065 64 98, 617 7, 190, 091 72.89
[3, 4) 98, 584 0.00050 49 98, 560 7, 091, 474 71.93
[4, 5) 98, 535 0.00040 40 98, 515 6, 992, 914 70.97
... ... ... ... ... ... ...
Looks like we shall need more notations...
Edward Furman Actuarial mathematics MATH 3280 2 / 15

Fractional ages
Definition 1.2 (Random number of survivors to age x.)

Let us have a group of l0 new born children. Then, for 1{j}
indicating the survival of the new born child number j to age x,
l0
X
L(x) := 1{j}
j=1
denotes the number of children alive at age x. L(x) is an r.v.
Definition 1.3 (Expected number of survivors to age x.)

For x p0 = P[1{j} = 1] for every j = 1, . . . , l0 , the expectation of
L(x) is  
Xl0
lx := E[L(x)] = E  1{j} = l0 ·x p0 .
j=1
(Think of the binomial r.v.)

Fractional ages
Definition 1.4
Let n D(x) := L(x) − L(x + n) denote the group of deaths
between ages x and x + n. We then define the expected
number of deaths (out of l0 and between the aforementioned
ages)
n dx := E[n D(x)] = l0 (x p0 − x+n p0 ) = lx − lx+n .
Proposition 1.1
We have that
1 d
µ(x) = − · lx .
lx dx
Proof.
Noticing that x p0 = lx /l0 completes the proof.

Fractional ages
At home.
Check that
Z x+n
lx+n = lx · exp − µ(s)ds ,
x
Z x+n
lx − lx+n = ls µ(s)ds.
x
Proposition 1.2
We have that the local extrema points of lx µ(x) correspond to
the points of inflection of lx .
Proof.
d d 1 d d2
lx µ(x) = − lx lx = − 2 lx ,
dx dx lx dx dx
which completes the proof.
Fractional ages
Figure: Plot of lx

Fractional ages
Figure: Plot of lx µ(x )

Fractional ages
Figure: Plot of µ(x )

Fractional ages
Approximating life functions at fractional ages

Life table functions investigated hitherto specify the c.d.f.
of K (x) completely. To specify the c.d.f. of T (x) we must
postulate an analytic form or adopt an assumption in
addition to the life table functions we have had.
We shall further review three different assumptions for
fractional ages, given a fixed x = 0, 1, . . . and t ∈ (0, 1),
1 Linear interpolation or the uniform distribution of deaths
(UDD),
S(x + t) = (1 − t)S(x ) + tS(x + 1).
2 Exponential interpolation or the constant force of mortality
(CFM),
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).

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

Fractional ages
The UDD
Linear approximation applied.

We find the value of lx+s , x = 0, 1, 2 . . . and s ∈ (0, 1) from the
following equations:
lx − lx+s x +s−x
= = s,
lx − lx+1 x +1−x
which yields
lx+s = lx − slx + slx+1 .
Finally, we find that:
lx+s = (1 − s)lx + slx+1 .

Fractional ages
In terms of the number of deaths, we have that

lx+s = (1 − s)lx + slx+1 = lx − sdx ,
where dx = lx − lx+1 .
The d.d.f.
Further, dividing by lx , we have that
UDD UDD
s px = 1 − sqx ⇔ s qx = sqx .
As qx is tabulated we can calculate s qx for any non-integer

duration s.
The p.d.f.
Also, we have that
d UDD
fT (x) (s) = s qx = qx , for 0 < s < 1.
ds
Fractional ages
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
The force of mortality.

Then the force of mortality is
UDD qx
µ(x + s) = ,
1 − s · qx
which increases in s.

Fractional ages
Age is fractional as well.

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

Fractional ages
Graphically, the main ideas of UDD can be seen as
Figure: lx+s decreases linearly and µx+s increases.

Constant force of mortality

The UDD approximation assumes a certain (linear) form of
t px , t ∈ [0, 1). Another way to assume specific behavior of
t px is to use its connection with the force of mortality
function. Namely:
CFM
For a fixed non-negative integer x and t ∈ [0, 1), let
µ(x + t) ≡ µ(x) for all t ∈ [0, 1). Then
Z t
CFM
t px = exp − µ(x + s)ds = exp(−t · µ(x)) = (px )t ,
0
and, we also similarly have that

!
Z 1
CFM
px = exp − µ(x + s)ds = e−µ(x) ⇔ µ(x) = − ln(px ).
0
Edward Furman Maths of Life Contingencies 1, MATH 3280 2 / 27

Both age and duration are non-integer

Fix 0 ≤ t < s < 1, then, for u = t + r ,
Z s−t
s−t px+t = exp − µ(x + t + r )dr
0
Z s
= exp − µ(x + u)du
Zt s
CFM
= exp − µ(x)du
t
= exp(−(s − t)µ(x)) = (px )s−t .
At home.
Due to the above, we have that only the duration is important,
i.e.,
s−t px+t = s−t px .
Why?
At home.
Show that the CFM is consistent with the formula for the
exponential interpolation.
Graphically, the main ideas of CFM can be seen as
Figure: lx+s decreases and µx+s is constant.

Proposition 1.1
Let J(u)(:= J) denote an r.v. representing the number of
fractional time units (u) lived during the year of death. Then
under the CFM the r.v.’s K and J) are generally dependent,
while they are independent under the UDD.
Proof.
By definition,
P[K = k, J < j] = P[k ≤ T < k + j] = P[k < T ≤ k + j]

UDD
= k pu · j qu+k = k pu · j · qu+k
= k | qu · j = P[K (u) = k] · P[J < k],
which completes the proof for the UDD case. For the CFM
case, j qu+k = 1 − j pu+k = 1 − (pu+k )j , which together with the
above, do not separate into a product.

Remark.
Note that if pu+k = pu , then k pu = (pu )k , and then
P[K = k, J < j] = P[k ≤ T < k + j] = P[k < T ≤ k + j]

CFM
= k pu · j qu+k = k pu · (1 − (pu+k )j )
= (pu )k · (1 − (pu+k )j )
(pu )k (1 − pu )(1 − (pu+k )j )
=
1 − pu
= P[K = k] · P[J < j].
Remember.
The UDD assumption implies that the r.v. J is uniform on [0, 1].
On the other hand, the CFM assumes that J is conditionally
exponential with µ, i.e., if X |X ≤ 1 ∽ Exp(µ), then J and X
have same distribution. Are the r.v.’s equal?

The Balducci assumption
Balducci approximation.
We have that, for 0 ≥ t < 1,
1 Hyp 1−t t (1 − t)lx+1 + tlx

= + = ,
lx+t lx lx+1 lx · lx+1
which implies that
Hyp lx · lx+1 lx+1

lx+t = = ,
lx+1 + tdx px + tqx
and hence, dividing by lx throughout,

Hyp px px
t px = = ,
px + tqx 1 − (1 − t)qx
and
1 − (1 − t)qx − px tqx
t qx = = .
1 − (1 − t)qx 1 − (1 − t)qx
Moreover, for 0 ≤ t < s < 1, we have that
s px Hyp 1 − (1 − t)qx
s−t px+t = = .
t px 1 − (1 − s)qx
as well as that
Hyp qx (s − t)
s−t qx+t = 1 − s−t px+t = .
1 − (1 − s)qx
The force of mortality can be found from

d Hyp d px
µx+t = − ln t px = − ln ,
dt dt 1 − (1 − t) qx
that is
Hyp d qx
µx+t = ln(1 − (1 − t) qx ) = ,
dt 1 − (1 − t) qx
which decreases in t.
Graphically, the main ideas of Balducci’s assumption can be
seen as
Figure: lx+s decreases and µx+s decreases.

To conclude the fractional ages
Remember.
For integer non-negative x and t and s in [0, 1), we have that
UDD CFM Balducci

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

Figure: Fractional ages’ concluding plot.

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 x
µ(x) = Bc , S(x) = exp − (c − 1) , B > 0, c > 1, x ≥ 0.
log(c)
Makeham (1867) used the following:

x B x
µ(x) = A + bc , S(x) = exp −Ax − (c − 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

Proposition 1.2
Let T (u) ∽ Fu . Then
◦
Z ∞ Z 1
−1
eu := E[T (u)] = t pu dt = F u (q)dq.
0 0
Proof.
The first integral follows by integration by parts, and for the
second one, and absolutely continuous F ,
Z ∞ Z ∞
−1
F u (x)dx = F u (x)d F u (F u (x))
0 0
Z 0 Z 0
−1 −1
= td F u (t) =− F u (t)dt
1 1
Z 1
−1
= F u (t)dt,
0
as required. Edward Furman Maths of Life Contingencies 1, MATH 3280 13 / 27

Proposition 1.3
For the set up as in the previous proposition, it holds that
Z ∞ ◦ 2
Var[T (u)] = 2 t t pu dt − eu .
0
Proof.
Integration by parts.
Definition 1.1
The median future lifetime of (u) is denoted by m(u), and it can
be found by solving:
F (u + m(u))
P(T (u) > m(u)) = = 0.5
F (u)
for m(u).
Definition 1.2
The mode for (u) can be obtained as:
arg max t px µ(x + t).

t∈R+
Proposition 1.4
Let K (u) ∽ Fu . Then

∞
X
eu := E[K (u)] = k pu .
k =1
To prove the latter identity, first define the forward difference

operator by:
∆f (x) = f (x + 1) − f (x).

Remember.
And second recall the summation by parts formula:
n
X n
X
n+1
f (k)∆g(k) = [f (k)g(k)] |m − g(k + 1)∆f (k).
k =m k =m
Proof of Proposition 1.4.

Then:
∞
X
eu = kP(K (u) = k)
k =0
X∞
= k (P(K (u) ≤ k) − P(K (u) ≤ k − 1))
k =0
X∞
= k (k +1 qu − k qu ) .
k =0

Proof.
Further
∞
X ∞
X
eu = k∆k qu = k∆(1 − k pu )
k =0 k =0
X∞
= − k∆k pu .
k =0
And hence, using the summation by parts:

∞
!
X
0 −
eu = − k k pu |∞ k +1 pu ∆k
k =0
∞
X ∞
X
= k +1 pu = k pu ,
k =0 k =1
bearing in mind that eu is finite, that completes the proof.

Proposition 1.5
Show that under the UDD assumption, we have that:

◦
eu = eu + 0.5.
Proof.
1X 1
Z ∞ Z
◦ 1 ∞
eu = lu+t dt = lu+k +t dt
lu 0 lu 0
k =0
1X 1
∞ Z
UDD
= ((1 − t)lu+k + tlu+k +1 ) dt
lu 0
k =0
∞
1X 1 1
= lu+k + lu+k +1
lu 2 2
k =0
∞
!
1 1 X 1
= lu + lu+k = + eu
lu 2 2
Edward Furman k =1 Maths of Life Contingencies 1, MATH 3280 18 / 27
Important note.
The fact that the UDD holds for (x) does not necesserily imply
same on (u).
Proof.
To UDD holds on (u) iff for k = 0, 1, 2, . . . and t ∈ [0, 1),
k +t pu = (1 − t)k pu + tk +1 pu
⇔ k pu t pu+k = (1 − t)k pu + tk pu pu+k
⇔ t pu+k = 1 − t qu+k = 1 − t(1 − pu+k ) = 1 − tqu+k
⇔ t qu+k = tqu+k
that is true under the UDD. Set (u) := (x : n ), then if the UDD
holds for (u), it must be
UDD
k +t px:n = (1 − t)k px:n + tk +1 px:n ,
which is not true, for k = n − 1, we have 0 6= (1 − t)n−1 px .

Important.
Note that Proposition 1.5 relates between the fully continuous
and fully discrete total expectancy of life. For instance, it
connects between the expectation of the r.v.’s min(T (x), T (n ))
and
⌊min(T (x), T (n )⌋ = min (⌊T (x)⌋, ⌊T (n )⌋) = min(K (x), n − 1).
Importantly, the expectation of the latter r.v. is

n−2
X n−1
X
E[min(K (x), n − 1)] = k k | qx + (n − 1)n−1 px = k px .
k =0 k =1

Proposition 1.6
We have that
∞
X
Var[K (u)] = (2k − 1)k pu − eu2 .
k =1
Proof.
We have that
∞
X
2
E[K (u) ] = k 2 k pu qu+k
k =0
X∞
= − k 2 ∆(k pu )
k =0
∞
X
= −k 2 k pu |∞
0 + k +1 pu ∆k
2
k =0

Proof.
∞
X
2
E[K (u) ] = (2k + 1)k +1 pu
k =0
X∞
= (2k − 1)k pu ,
k =1
which completes the proof.

Definition 1.3
The function Tu is defined as:
Z ∞ Z ∞
Tu := lu+t dt = ly dy,
0 u
and it represents the expected total future lifetime of a group of

lu individuals all aged u.
To grab an intuition, recall that:

Z ∞ Z
◦ ∞
lu+t
eu = t pu dt = dt.
0 0 lu
Hence, the following motivates the definition

◦
Tu = lu eu .

Definition 1.4
The function Lu is defined as:
Z 1 Z u+1
Lu := lu+t dt = ly dy,
0 u
and it denotes the total expected time lived between ages u

and u + 1 by lu lives aged u.
To grab an intuition, note that
Lu = Tu − Tu+1 .

Definition 1.5
The central rate of mortality is defined as:
R1
0 lu+t µ(u + t)dt du
mu := R1 = ,
Lu
0 lu+t dt
which is the weighted average force of mortality experienced

between ages u and u + 1 with weight lu+t at age u + t.
To get the right most side, divide both numerator and

denumerator by lu , and recall that
Z 1
t pu µ(u + t)dt = qu ,
0
and Z 1
tpu = Lu .
0
Proposition 1.7
◦
The following recursion holds for f (u) being eu and eu :
c(u) 1
f (u) = c(u) + d (u)f (u + 1) and f (u + 1) = − + f (u).
d (u) d (u)
Proof.
Note first that
∞
X ∞
X ∞
X
eu = k pu = pu + k pu = pu + pu k pu+1 = pu + pu eu+1 ,
k =1 k =2 k =1
◦
thus c(u) = d (u) = pu , u(∞) = 0. In the case of eu , and in the
fashion similar to the above one, we have that
Z ∞ Z 1 Z ∞
◦
eu = t pu dt = t pu dt + t pu dt
0 0 1
Proof.
◦
Z 1 Z ∞ Z 1 ◦
eu = t pu dt + pu t pu+1 dt = t pu dt + pu eu+1 ,
0 0 0
and thus Z 1
c(u) = t pu dt and d (u) = pu .
0
Remark.
If in addition, we assume linear lu+t for t ∈ [0, 1], then we
approximate c(u) above as
Z 1
UDD 1 1 1
c(u) = (1−tqu )dt = 1− qu = 1− (1 − pu ) = (1 + pu ) .
0 2 2 2

Lec2 PDF

Uploaded by

Copyright:

Available Formats

Lec2 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lec2 PDF

Uploaded by

Copyright:

Available Formats

Life tables with one source of decrement

Definition 1.1 (Life table.)

Looks like we shall need more notations...

Edward Furman Actuarial mathematics MATH 3280 2 / 15

Definition 1.2 (Random number of survivors to age x.)

denotes the number of children alive at age x. L(x) is an r.v.

Definition 1.3 (Expected number of survivors to age x.)

(Think of the binomial r.v.)

n dx := E[n D(x)] = l0 (x p0 − x+n p0 ) = lx − lx+n .

Edward Furman Actuarial mathematics MATH 3280 4 / 15

Edward Furman Actuarial mathematics MATH 3280 6 / 15

Figure: Plot of lx µ(x )

Edward Furman Actuarial mathematics MATH 3280 7 / 15

Figure: Plot of µ(x )

Edward Furman Actuarial mathematics MATH 3280 8 / 15

Approximating life functions at fractional ages

Edward Furman Actuarial mathematics MATH 3280 9 / 15

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

Edward Furman Actuarial mathematics MATH 3280 10 / 15

Linear approximation applied.

lx+s = (1 − s)lx + slx+1 .

Edward Furman Actuarial mathematics MATH 3280 11 / 15

In terms of the number of deaths, we have that

As qx is tabulated we can calculate s qx for any non-integer

As fT (x) (s) is constant in s and equal to qx , deaths are said

We have seen that

The force of mortality.

Edward Furman Actuarial mathematics MATH 3280 13 / 15

Age is fractional as well.

which after applying the UDD assumption reduces to

Edward Furman Actuarial mathematics MATH 3280 14 / 15

Graphically, the main ideas of UDD can be seen as

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

Edward Furman Actuarial mathematics MATH 3280 15 / 15

Constant force of mortality

and, we also similarly have that

Edward Furman Maths of Life Contingencies 1, MATH 3280 2 / 27

Both age and duration are non-integer

Graphically, the main ideas of CFM can be seen as

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

Edward Furman Maths of Life Contingencies 1, MATH 3280 4 / 27

P[K = k, J < j] = P[k ≤ T < k + j] = P[k < T ≤ k + j]

Edward Furman Maths of Life Contingencies 1, MATH 3280 5 / 27

P[K = k, J < j] = P[k ≤ T < k + j] = P[k < T ≤ k + j]

Edward Furman Maths of Life Contingencies 1, MATH 3280 6 / 27

The Balducci assumption

1 Hyp 1−t t (1 − t)lx+1 + tlx

which implies that

Hyp lx · lx+1 lx+1

and hence, dividing by lx throughout,

The force of mortality can be found from

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

Edward Furman Maths of Life Contingencies 1, MATH 3280 9 / 27

To conclude the fractional ages

UDD CFM Balducci

Edward Furman Maths of Life Contingencies 1, MATH 3280 10 / 27

Figure: Fractional ages’ concluding plot.

Edward Furman Maths of Life Contingencies 1, MATH 3280 11 / 27

De Moivre (1725) used a one parameter formula: