Stat 476

Life Contingencies II

Multiple Life and Multiple Decrement Models

Multiple Life Models

To this point, all of the products we’ve dealt with were based on
the life status (alive / dead or multi-state) of a single individual.

However, in order to deal with products involving multiple lives, we

need more complex models.
These future lifetimes may be dependent or independent.

The force of transition from state i to state j based on the joint life
statuses of (x) and (y ) is denoted by µijxy or µijx:y , with the latter
being preferred when we have numerical values for x and/or y .

2
Joint Life Model

(x) alive - (x) alive

(y ) alive (y ) dead

0 1
PP
PP
PP
P PP
? q ?
(x) dead - (x) dead
(y ) alive (y ) dead

2 3

The model above could be used in cases where cash flows depend
on the status of two lives (x) and (y ). If the model allows a
transition from state 0 directly to state 3, it’s known as a
common shock model.

3
Traditional Multiple Life Status Actuarial Notation
The subscript xy denotes a status which fails upon the first of the
future lifetimes of (x) and (y ) to fail. Likewise, the status xy fails
upon the failure of the second failure of (x) and (y ).
We will use these statuses in the same manner we used the
single life status x and the term status n
This notation follows the same convention we used earlier for
term products and guaranteed annuities.

Then Txy is the random variable describing the time until the first
death of (x) and (y ), and Txy is the RV giving the time until the
second death of (x) and (y ).

Important Relationship
Tx + Ty = Txy + Txy

4
Multiple Life Probability Definitions and Relationships

µxy = force of mortality for the joint status xy

= µ01 02 03
xy + µxy + µxy

t pxy = P[the status xy does not fail within t years]

= P[(x) and (y ) are both alive in t years]
00
= t pxy

t qxy = P[the status xy fails within t years]

= 1 − t pxy
= P[(x) and (y ) are not both alive in t years]
01 02 03
= t pxy + t pxy + t pxy

5
Multiple Life Probability Definitions and Relationships

t pxy = P[the status xy does not fail within t years]

= P[(x) or (y ) or both are still alive in t years]
00 01 02
= t pxy + t pxy + t pxy

t qxy = P[the status xy fails within t years]

= 1 − t pxy
= P[(x) and (y ) are both dead in t years]
03
= t pxy

We have the following relationship for (x) and (y ):

t px + t py = t pxy + t pxy

Further, we define n Exy = v n n pxy and n Exy = v n n pxy

6
More Multiple Life Probability Definitions and
Relationships
We also use the numbers 1, 2, ... above the individual life statuses
to indicate a specific order of status failure:

t qxy
1 = P[(x) dies before (y ) and within t years]
Z t
00 02
= s pxy µx+s:y +s ds
0

t qxy
2 = P[(x) dies after (y ) and within t years]
Z t
01 13
= s pxy µx+s ds
0

Note that we can let t be ∞ in either of these symbols.

7
Insurances and Annuities Based on Multiple Lives

We can derive some relationships among the various multiple life

insurances and annuities:

Axy + Axy = Ax + Ay axy + axy = ax + ay

1 − Axy 1 − Axy
axy = axy =
δ δ

ax|y = ay − axy ax|y + ay |x = axy − axy

8
Two Independent Lives
In the special case where we’re dealing with two future lifetimes
that are independent, many of our expressions become much
simpler. In this event, all of the forces of transition only depend on
a single life, so in our joint life model, we would have:

µ01 23
xy = µy = µy µ02 13
xy = µx = µx µ03
xy = 0

As a result, our transition probabilities become functions of the

individual survival probabilities:
00 01 00 01
t pxy = t px t py t pxy = t px t qy t pxy = t px t py t pxy = t px t qy

And for the case of independent lives, in the traditional actuarial

notation, we have:

µxy = µx + µy and t pxy = t px t py

9
Two Independent Lives – Example

Suppose that (x) and (y ) have independent future lifetimes with

µx = 0.02, µy = 0.03, and δ = 0.05.

(a) Calculate ax|y .

Z ∞ Z ∞
−δt
ax|y = e 02
t pxy dt = e −δt t qx t py dt = 2.50
0 0

(b) Calculate the probability that (x) dies before (y ).

Z ∞ Z ∞
00 02
∞q 1 = t pxy µx+t:y +t dt = t px t py µx+t dt = 0.4
xy 0 0

10
Multiple Decrement Models

Another common application of multi-state model theory is in

situations where a members of a population may leave due to one
of several different causes. These so-called multiple decrement
models can applied to various situations.

These models typically have one active / alive state, where

each individual begins. An individual may leave the population
by transitioning to any of the exit states.

We usually only model a single transition out of the active

state; we either ignore or disallow transitions among the exit
states.

11
Multiple Decrement Model

Exit 1
-
Active

0 PP 1
PP
P PP
P PP
PP
P P
q
P
Exit 2

2
? ...
Exit n

12
Multiple Decrement Example

We issue a fully continuous whole life policy to (x). The death

benefit is $100,000, but doubles if (x) dies as a result of an
accident. No benefit or surrender value is payable if the
policyholder lapses his policy.

(a) Draw an appopriate state diagram for modeling this type of

policy.

(b) Write an expression for the probability that this policy ever
pays a benefit.

(c) Write an expression for the EPV of this insurance.

13
Multiple Decrement Tables and Traditional Actuarial
Notation
The probabilities for multiple decrement models are traditionally
expressed in multiple decrement tables, which are analogous to
the mortality tables used in our alive-dead model.
We again start with some cohort of `x0 people at age x0 . The
number of people remaining in the population at age x is
given by `x .
The number of people leaving at age x due to decrement i is
(i)
denoted by dx ; the total number leaving during this year is
(τ )
denoted by dx , where
(τ )
X (i)
dx = dx
i

(1) (2) (n) (τ )

Then `x+1 = `x − dx − dx − · · · − dx = `x − dx

14
Sample Double Decrement Table

Here’s an excerpt of a table with two decrements:

(1) (2)
x `x dx dx
60 1000 11 10
61 979 12 10
62 957 13 10
63 934 14 10
64 910 15 10
.. .. .. ..
. . . .

We can calculate that of the original 1000 members of this

(2)
population, d2 = 10 of them leave due to decrement 2 in the
(τ ) (1) (2)
third year, and d3 = d3 + d3 = 24 of them leave in the fourth
year for any reason.

15
Multiple Decrement Tables — More Notation

We denote the probability of leaving in a given year as a result of

decrement i by
(i)
(i) dx
qx =
`x
and the probability of leaving due to any decrement as
(τ )
(τ )
X (i) dx
qx = qx =
`x
i

We denote the probability of surviving all decrements in a given

year by
(τ ) (τ ) `x+1
px = 1 − qx =
`x

16
Multiple Decrements in Continuous Models
In a continuous model, the force of failure for decrement i at age x
(i)
is denoted by µx and the total force of failure is
(τ ) (1) (2) (n)
µ x = µx + µ x + · · · + µx

The probability of failing within t years due to decrement i is:

Z t
(i) (τ ) (i)
t qx = s px µx+s ds
0

and the probability of failing within t due to any decrement is:

Z t
(τ ) (τ ) (τ )
t qx = s px µx+s ds
0

(Note that these aren’t actually new formulas; they’re just

expressed in different notation.)

17
Dependent and Independent Probabilities for Multiple
Decrement Models
(i)
In our multiple decrement table above, the dx values (and hence
(i)
the calculated qx values) depended on the other decrements in
(i)
the table; hence, qx is sometimes called a dependent probability
of decrement.

If we calculate the probability of someone age x failing due to

decrement i in the absence of other decrements, the resulting
probability is called the independent probability of decrement,
the probability of decrement in the associated single
decrement table, or sometimes the absolute rate of decrement
0(i)
and is denoted by qx .
0(i) (i)
Note that t qx ≥ t qx for any x, t, and i.

18
Formulas for Independent Probabilities in Multiple
Decrement Models

The probability of x still being in the population in t years in the

associated single decrement model for decrement i is denoted
0 (i)
t px .

n
0 (i) 0 (i) (τ ) 0 (i)
Y
t px =1− t qx t px = t px
i=1

 Z t   Z t 
0 (i) (i) (τ ) (τ )
t px = exp − µx+s ds t px = exp − µx+s ds
0 0

19
Fractional Age Assumptions in Multiple Decrement Models

If we have a discrete model (i.e., a multiple decrement table), we’ll

often need to make some assumptions regarding how the
decrements act within the year:
We need this in order to get the independent failure
probabilities from the dependent ones (and vice-versa).
We also need this additional information to do calculations for
non-integral lengths of time.

20
Fractional Age Assumptions Example
(Decrement 1 is retirement, decrement 2 is death)
(1) (2)
x `x dx dx
60 1000 11 10
61 979 12 10
62 957 13 10
63 934 14 10
64 910 15 10
.. .. .. ..
. . . .

Example: For the double decrement table above, calculate and

0 (1) 0 (2)
interpret q62 and q62 assuming:
(a) All retirements occur at the beginning of the year.
(b) All retirements occur at the end of the year.
(c) All retirements occur uniformly throughout the year.

(Assume in all cases that deaths happen throughout the year.)

21
Uniform Distribution of Decrements in the
Multiple-Decrement Table
One option is to assume that each decrement in the multiple
decrement table is distributed uniformly over the year in the
(j)
presence of other decrements, i.e., t qx is a linear function of t.

Under this UDD assumption in the multiple decrement context:

(j) (j)
t qx = t · qx

From this assumption, we can derive the following relationships:

(j)
(τ ) (τ ) (τ ) (τ ) (j) qx
t qx = t · qx t px = 1 − t · qx µx+t = (τ )
1 − t · qx
 (j) (τ )
(τ ) qx /qx

0(j)
t px = 1 − t · qx

22
Uniform Distribution in the Associated Single-Decrement
Tables
Another option is to assume that each decrement is distributed
uniformly in its associated single decrement table, i.e., in the
0(j)
absence of other decrements, so that t qx is a linear function of t.

Under this UDD assumption in the single decrement context:

0(j) 0(j)
t qx = t · qx

From this assumption, we can derive various relationships. For

example, for a double decrement table, we have:
   
(1) 0(1) 1 0(2) (2) 0(2) 1 0(1)
qx = qx 1 − · qx and qx = qx 1 − · qx
2 2

Similar equations can be derived for triple decrement tables.

23
Constant Force of Decrement Assumption

We could instead assume that the forces of decrement are constant

within the year. This assumption results in the following formulas:

0(j) t (τ ) t
h i h i
0(j) (τ )
t px = px t px = px

(j)  i
(j) qx h
(τ ) t
t qx = (τ )
1 − px
qx

i (j) (τ )
(τ ) qx /qx
h
0(j)
px = px

24
Multiple Decrement Examples I
This table is from a policy which expires on first of death (1),
surrender (2), or illness (3). Calculate

(1) (2) (3)

x `x dx dx dx
40 100000 51 4784 44 (τ ) (1) (3)
(a) 3 p45 , q40 , and 5 q41 .
41 95121 52 4526 47
42 90496 53 4268 50 (b) the probability that a policy
43 86125 54 4010 53 issued to (45) generates a claim for
44 82008 55 3753 56 death or critical illness before age 47.
45 78144 56 3496 59
46 74533 57 3239 62 (c) the probability that a policy
47 71175 57 2983 65 issued to (40) is surrendered between
48 68070 58 2729 67 ages 45 and 47.
49 65216 58 2476 69
50 62613 58 2226 70

25
Multiple Decrement Examples II
This table is from a policy which expires on first of death (1),
surrender (2), or illness (3). Calculate

(1) (2) (3)

x `x dx dx dx
40 100000 51 4784 44
41 95121 52 4526 47 (j)
Calculate 0.2 q50 for j = 1, 2, 3
42 90496 53 4268 50
assuming, between integer ages,
43 86125 54 4010 53
44 82008 55 3753 56 (a) UDD in all decrements
45 78144 56 3496 59
46 74533 57 3239 62 (b) constant transition forces in all
47 71175 57 2983 65 decrements
48 68070 58 2729 67
49 65216 58 2476 69
50 62613 58 2226 70

26