Professional Documents
Culture Documents
Decr Ement
Decr Ement
MULTIPLE DECREMENTS
April 20 2016
duncan@pstat.ucsb.edu
1
MULTIPLE DECREMENTS
Life Insurance
death
withdrawal (lapse)
surrender (termination with refund of accumulated value)
etc.
Pensions
Health Insurance
2
Because health insurance is priced on an annual basis and paid for monthly,
decrements play a lesser role in pricing. However, in health insurance the
reason that someone terminates from a group is very important. Why do you
think that may be?
Terminology
Forces that reduce a population (insured lives; annuitants; etc.) are referred to as
forces of decrement.
You will remember (from 172A) that the probability that a member of a
population subject to a single force of decrement (e.g. mortality) stays in force for
t years is t px . Analogously, the probability that a member of a population
subject to multiple forces of decrement stays in force for t years is t px ( ) .
We denote the probability that (x) survives t years under multiple decrements as
t px ; (Greek t) "total."
( )
Recall that, in the case of a single decrement constant force model (e.g.
t
x s ds
mortality): t px e 0
(Result 2.19 from 172A)
3
Introducing multiple decrements, for example, let us consider a simple pension
plan model that recognizes only two decrements: death and retirement, with
forces of decrement µx(r) and µx(d) respectively. (r = retirement; d = death)
EXAMPLE 1:
SOLUTION
t
x( s) ds
n
Generalizing:
( )
x t ( j)
x t and survival probability, t px e 0
j 1
!
t
-0
e
d
s
μ
s
)
(
1
)
(
4
1
)
(
1
And of course, t q
!
x s p! x (1) x(1) s ds = 1 t p! x (1)
p
0
1x
)
(
You may also see referred to as the net probability of decrement, the
t
!
t
( )ds
This follows because: t px e 0
or said another way, the total survival probability (surviving the combined effect
of death and retirement) is the product of the probability that (x) survives death
only for t years times the probability that he survives retirement only for t years.
Of course, in pension plans, t px! ( 2) 1.0 for t < the first age of retirement
eligibility, frequently 55. So be aware that decrements are not always positive
(for some range) or uniform.
in other words (fairly obviously) the rate of decrement is higher when the
decrement acts alone than when it acts in combination with other decrements.
Since qx! (i ) qx(i ) ,px! (i ) px(i )
5
Note (3) that: qx ( ) qx (1) qx ( 2)
j
n n
( )
t px exp ( 0
t
µ x s ds
( j)
t px ! ( j)
The use of primed and unprimed probabilities can be confusing. Useful to keep in
mind that the sum of the unprimed values is equal to the total value (). If you
are dealing with unprimed values, you need to convert to primed or use the
relationship:
(Primed) px . t px t px ( ) (Unprimed)
! r ! d
t
6
An example to illustrate primed/un-primed (acting alone/acting together)
probabilities
100
90
x X+1
When the decrements act on their own the number dying is qxd 100 or 10 and the
number withdrawing is qxw 100 or 40. When the two act together, however,
some of the lives who would otherwise die have withdrawn (and vice-versa) so
when the two decrements act together their effect is less than it would be if they
acted alone. The number dying is qx (80) or 8 and withdrawing is qw (95) or 38.
Thus:
7
For this example we assumed UDD for simplicity; from 172A you know that UDD is
not the only assumption about mortality during year-of-age x; we will see more in
a moment.
EXAMPLE 2
Calculate qx! 2
SOLUTION
qx
! 1
qx! ( 2 ) qx ( ) 0.18 1 p x
( )
0.18
1 – 1 qx! (1) 1 qx! ( 2 ) 0.18
q
x
! (1)
q x
! (2)
– q q
x
! (1)
x
! ( 2)
0.18
q q
x
! (1)
x
! ( 2)
0.18
2 q q x
! (1)
x
! (1)
0.18
q 0.3
x
! (1)
8
Joint Distributions
We define a new random variable, Jx that denotes the mode of decrement at age
x. The force of decrement is the instantaneous rate of decrement due to a
specific cause:
This expression means: the probability that the future lifetime of x lies between t
and t+dt, given that the lifetime of x is terminated by decrement j.
n
And: µ xt ( )
j
µ x t ( j )
And t qx 0 px ( ) µ xs ( j ) ds
t
( j)
s (Important result)
EXAMPLE 3
Calculate qx ( 2)
9
SOLUTION
e 0
µ (1) s ......
µ x( n ) s .ds
Survival probability: t px (t ) x
! j
and t px (t ) j t px
f t , i t px (t ) .µ x t
i
10
The total probability of termination (all cause) is:
qx ( ) t qx ( )1 t qx ...... n
(1 t px ( ) )
2
t t qx
i
0 f s, i . ds lim t qx
i
qx =
t
11
More things to remember:
1. t px ( ) i t px
! (i ) ( t px ( ) is the product of the probabilities acting alone)
2. t qx ( ) i q
t x
(i ) (note; sum of the t qx(i ) , acting together)
When no assumption is stated (such as UDD or Constant Force) the question will
probably want you to use an integral such as the first relationship in the box.
When an assumption is stated it will be one of the forms UDD or CFM.
EXAMPLE 4
SOLUTION
Also, the density function for the remaining lifetime of x is (as we have seen
before) t px µx+t
t
e 0.06 s 0.01
. ds
0
So, substituting t = 1:
q x(d) = 1 (1 - e -0.06) = 0.00971 (note < q x!(d) see below for derivation)
6
(d)
Finally: tq x = 1 (1 - e -0.06t) and t q x(r) = 5 (1 - e -0.06t)
6 6
13
As noted previously,
This is an important example; take a minute to grasp it because we will see how
you can derive the same result assuming Uniform Distribution of Decrements
(UDD) as follows.
We found the relationship between q x (r) and q x!(r) by integration; we could also
have found this by using the formula:
EXAMPLE 5
14
SOLUTION
20 t 30 t 1
q ( 2)
56 1/ 20 t dt 30t – t 2 / 2 |56
600
51
| 10
600
= 49/1200
b. For the second part of the question, we want Pr(J10 = 1), i.e. the probability
that (10) terminates from cause 1.
15
EXAMPLE 6
Find similar results for a 3-decrement model with constant force of decrement.
Find ∞ tqx (j) , j = 1,2,3. Forces of decrement are:
Hint: first calculate survival functions for each decrement acting alone. Then
develop density functions for the decrements.
SOLUTION
()
tp x = tp x!(1) tp x!(2) tp x!(3) = e-0.08t
16
!(1)
RELATIONSHIP BETWEEN tq x and t q x(1)
Until now we have moved between t q x!(1) and t q x(1) using the relationship
t n
( )
µ x s ds
n
t px exp ( j)
p!x ( j )
t
0 j
As a result we can form a relationship between primed and unprimed rates using
p ' s and p! ' s.
This assumption results in the following formula (in the case of a double-
decrement model; we shall generalize this in a moment):
!( 2 ) !(1)
q q
qx (1)
qx !(1)
(1 x ) and conversely: qx (2)
qx !(2)
(1 x )
2 2
Take a look back at the graph on page 7; these formulae are a formulaic
equivalent of that graphic.
17
EXAMPLE 7
We assume that decrements are uniformly distributed over the year-of-age in the
associated single decrement table. Calculate lx(τ) .
SOLUTION
qx ( 2) d x ( 2) / lx qx! ( 2) 1 – ½ qx!(1)
q20!(2) = 0.01264
18
Constant force of mortality
px ( ) px ( )
t
a. t (allows us to calculate survival probability for fractional
ages/durations).
µ x s
j
b. qx ( j ) for s in the interval [0, 1)
t
t qx
µ x s ( )
We use this relationship to get unprimed rates from the aggregate (sum of
primed)
qx( j ) / qx( )
c. t px
! ( j)
t px ( )
EXAMPLE 8
Assume a constant force of decrement for each of 3 decrements over each year of
age in a triple decrement model. You are given
a. q50(1) = q50(3)
b. q50(2) = 2 q50(1)
c. µ50+t(1) = log (2) for 0< t< 1
SOLUTION
19
We are given µ50(1) = log (2) and can derive µ50(2) as 2 log 2 or log 4. (Applying
µ x s
j
result b. above qx ( j ) which says that the µ’s are in same ratio as
t
t qx
µ x s ( )
the qx’s.)
= (1 – ¼) (2/4) = 3/8
20
Discrete Decrements
These decrements tend to be tricky but with a clear head you should be ok. Here
is an example from an SOA exam:
EXAMPLE 9
Calculate q 61(1)
SOLUTION
21
q61
1
For later use, !1
p61!(1) µ (1) 61t dt which is just a re-statement of a result
0 t
from 172A:
We can determine p 61!(2) quite simply by realizing that if the decrement takes
place at the end of the year, the probability of surviving to the end of the year
from the single decrement (2) is 1.0.
Similarly for p 61!(3) , we realize that all the decrements take place at the
beginning of the year and that the probability of surviving a year from
decrement 3 is simply 0.8.
= 0.074
Note: this is not an application of the formula for the triple decrement; you are
not given an assumption about the distribution within the year of age and we are
given a force of decrement (µ 61+t) so you should use the integral. Students given
this type of problem in an exam often make the mistake of applying UDD but
there is nothing in the question that implies this.
There is a formula that you can apply in these situations (as we discussed in class).
22
CONSTRUCTION OF TABLES
Large clients may have enough data to construct a multiple decrement table from
their own experience. This is rarely the case. More likely a table will be
constructed using associated standard single decrement tables. Either way, the
process will probably involve modifying or scaling an existing table. So in pension
plan valuation (for example) you may have enough mortality experience to
aggregate this over a period of years to evaluate how your actual data compares
with expected mortality from a standard table. This can get quite tricky because
(of course) the observed deaths are qx, not qx! , thus requiring some additional
allowance in comparing with standard (single decrement) mortality tables.
EXAMPLE 10
Assume a radix of 10,000 and a double decrement table beginning at age 65;
q65!(1) = 0.01; q65!(2) = 0.02. Find l 66 () .
SOLUTION
23
q65(2) = q65!(2) (1 - q65!(1)/2) = 0.02 * 0.995 = 0.0199
l 65 () = 9,702
For the purpose of this course, you will need to understand the application of
multiple decrements theory to:
Good news: this will all be a logical extension of what you have already seen in
the single decrement world.
As you know, in the case of a continuous annuity āx v t t px dt
0
We can generalize this in the multiple-decrement case to: āx v t t px( ) dt
0
Similarly, for an Insurance policy (single decrement): Ā x v t t px µx t dt
0
If benefits are always the same, irrespective of the cause of death, the formula
generalizes to: Ā x v t t px( ) µx( dt) dt where d= death.
0
24
However, benefits often differ according to the cause of decrement. The primary
example of this is a type of policy called “Double Indemnity” (if you haven’t seen
the famous movie “Double Indemnity” you should rent it (the original 1944
version with Barbara Stanwyck and Fred McMurray). While we are on the topic
of movies, you should also watch the other famous actuarial movie, “Billion Dollar
Bubble” which is the true story of a scam perpetrated by actuaries in an insurance
company in LA. I gave a copy to the club some years ago – ask Sarah. Otherwise
there is an excerpt on Youtube).
Anyway, back to Double Indemnity policies. The idea behind Double Indemnity
policies is that they pay the sum insured if you die from natural causes, but if you
die as a result of an accident they pay an Accidental death benefit. So in the
case of death from an accidental cause the policy pays twice (Double Indemnity).
There are also Accidental death only policies which pay only on death due to
accident. When pricing Accidental Death policies, of course, we have to take
account of all causes of death, not just death by accident, although only in the
case of death by accident is a benefit payable. There are also disability policies
that work like this, called Total and Partial Disability. Total disability entitles you
to a benefit (usually an annuity) while there are specified payments for loss of
function: so much for a finger, so much for a hand, etc. (Yes, rather grisly.)
n
APV 0
j 1
bt ( j ) vt t px ( ) µ xt ( j ) dt.
25
EXAMPLE 11
26
27
EXAMPLE 12
SOLUTION
28
EXAMPLE 13
What are the benefit reserves for this policy at the end of year 1 and year 2?
SOLUTION
29
Asset Shares
We have seen before how to calculate surplus (or gain/loss) on the policy:
30
31
EXAMPLE 14
SOLUTION
32
Non-forfeiture Options (more on this under Profit Testing in the next lecture)
( k Wx:n| )(Axk:nk| ) (APV of the RPU amount) = Reserve (or cash value)
Px:n|
1 (Since Ax+k: n k | = P x+k:n-k ä x+k: n k | )
Pxk:nk|
33
Exam questions about Non-forfeiture values frequently involve a surrender
charge. This is generally of the form (α)( kVx: n | ) so the solution is simple as long as
you remember to deduct the surrender charge from the amount available to the
policyholder.
Using our prior (Endowment Insurance) example with cash value at duration
k = denoted kCV and a reduced paid-up (we usually abbreviate to RPU) face
amount of b:
kCV = b A !x+k: n k |
It is important to note that the policy has the same duration as before (n-k years
remaining). Any change in underlying policy terms such as duration could give
rise to additional risk and should be priced into the conversion. One situation
that can arise is that the cash value is large enough (and n-k short enough) that
the face amount could exceed the prior sum insured. Clearly this is anti-selective
on the part of the policyholder. One way to address this is to maintain the face
amount at 1 and to pay remaining cash value out as a pure endowment.
EXAMPLE 15
A life aged 40 buys a whole life insurance of 1000 payable at the moment of
death. Premiums are payable continuously until death.
a. Calculate the net level premium for the policy and the benefit reserve at
duration 15.
b. A surrender charge of 10% of reserve is applied. If the life converts at
duration 15 to a Reduced paid-up insurance, calculate the amount of the
RPU.
c. Suppose, instead, that the life elects a 20-year Extended Term insurance.
What is the face amount of the extended term insurance?
34
SOLUTION
35
c. We need Ā! 55: 20 | = _1_ (1 – vn) = (1/45)(1- 0.3769)
ω-x δ 0.05
= 0.2769
A quick check shows that this is not more than the prior face amount (1000)
so there is no selection involved.
36