Documente Academic
Documente Profesional
Documente Cultură
Yeng M. Chang
2013
All Rights Reserved
2
Contents
Preface 7
3 Life Tables 35
3.1 Calculating Probabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Fractional Age Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.1 Uniform Distribution of Deaths (UDD) Assumption . . . . . . . . . . . . . . 38
3.2.2 Constant Force of Mortality (CF) Assumption . . . . . . . . . . . . . . . . . . 40
3.3 Select and Ultimate Life Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4 SOA Released Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.5 Formula Summary Sheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3
4 CONTENTS
4 Valuation of Insurances 45
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Types of Insurances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 Variances and Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.1 Discrete Whole Life Insurance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.2 Continuous Whole Life Insurance . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.3 Term Insurances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3.4 Deferred Insurances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3.5 Pure Endowment Insurance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.3.6 Endowment Insurances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.4 Mortality Assumptions on the Entire Domain of Tx . . . . . . . . . . . . . . . . . . . 52
4.4.1 Continuous Insurances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.4.1.1 Fundamental Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.4.1.2 Whole Life Insurance (Ax ) . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4.1.3 Term Insurance (A1xn ) . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4.1.4 Deferred Whole Life Insurance (n Ax ) . . . . . . . . . . . . . . . . . 54
4.4.1.5 Deferred Term Insurance (u A1xn ) . . . . . . . . . . . . . . . . . . . . 54
4.4.1.6 Endowment Insurance (Axn ) . . . . . . . . . . . . . . . . . . . . . . 54
4.4.1.7 Calculation of Higher Moments . . . . . . . . . . . . . . . . . . . . . 55
4.4.2 Discrete Insurances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4.2.1 Fundamental Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4.2.2 Whole Life Insurance (Ax ) . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4.2.3 Term Insurance (A1xn ) . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4.2.4 Deferred Whole Life Insurance (n Ax ) . . . . . . . . . . . . . . . . . 58
4.4.2.5 Deferred Term Insurance (u A1xn ) . . . . . . . . . . . . . . . . . . . . 58
4.4.2.6 Endowment Insurance (Axn ) . . . . . . . . . . . . . . . . . . . . . . 58
4.4.2.7 Calculation of Higher Moments . . . . . . . . . . . . . . . . . . . . . 58
4.5 UDD Assumption: Continuous-Discrete Relationships . . . . . . . . . . . . . . . . . . 59
4.6 mthly Insurances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.7 UDD Assumption: Discrete-mthly Relationships . . . . . . . . . . . . . . . . . . . . . 65
4.8 Insurances with Variable Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.8.1 General Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.8.2 Arithmetically Changing Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.9 Recursive Relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.10 Actuarial Accumulated Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.11 SOA Released Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.12 Formula Summary Sheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.13 Appendix: Fully-Variable Insurances . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5 Valuation of Annuities 77
5.1 Review from FM/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.1.1 Discrete Annuities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.1.2 Continuous Annuities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.1.3 mthly annuities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.3 Whole-Life Annuities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.4 Whole Life Annuity-Insurance Relationships and Moments . . . . . . . . . . . . . . . 82
5.5 Non-Insurance Annuity APV Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.6 kth Moments of Annuities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6 CONTENTS
Preface
This set of notes is meant to prepare you to take exam MLC by the Society of Actuaries. Numbers
surrounded by [ ] indicate a specific source cited in the bibliography, presented at the end of this
set of notes.
I base this set of notes on Models for Quantifying Risk by Cunningham, et al., (4th edition)
which I will abbreviate as MQR throughout this text (see [1]); Actuarial Mathematics for Life
Contingent Risks by Dickson, et al., which I will abbreviate as AMLCR (see [3]); their respective
supplementary notes (see [2], [4]); concepts which are in the SOA sample questions and released
exams for exam MLC; and my experiences from taking Math 450 and Math 460 at the University
of Wisconsin - Eau Claire with Prof. Herschel Day.
I will rely immensely on exam P and FM material as I go through this text. Information in footnotes
and appendices consist of technical information that is needed to add coherence to the discussion,
but is unnecessary for the the purpose of studying for the exam.
This set of notes would not exist without the help and support of many people. First of all, I thank
the professors whose courses I took as an undergraduate at the University of Wisconsin - Eau Claire.
Specifically, I would like to thank Dr. Carolyn Otto, Dr. Chris Ahrendt, and Dr. Dandrielle Lewis
for teaching me how to write mathematics and for pushing me to learn the LATEX typesetting
language. I would also like to thank Dr. Vicki Whitledge, my advisor and Probability professor
during my undergraduate years, as this text would not exist had I not taken her Probability course.
Also, I would like to thank Prof. Herschel Day for teaching me the course MLC material in Math
450 and Math 460.
I would also like to thank Jim Daniel, Abraham Weishaus, and Steve Paris for correcting errors,
helping out with proofs, and/or giving suggestions as I was creating these notes. In addition, I
would like to thank the various people who gave me feedback on my notes through the online
Actuarial Outpost.
Last, but definitely not least, I would like to thank my girlfriend Katie Miller for her support as I
took on this ambitious project during my undergraduate years.
I wish you the best of luck as you prepare for this exam! If you find any typos, mistakes, or have
any suggestions for this set of notes, please contact me at (e-mail address omitted).
Yeng Chang
7
8 CONTENTS
Chapter 1
FX (x) = P (X x) (1.1)
where P is the probability, hence P (X x) is read as the probability that X is less than or
equal to the value x. You may also see Pr(X x) instead of the above. Suppose that X is a
nonnegative continuous random variable, i.e., X 0. The CDF of X must satisfy the following
properties:
1. FX is continuous and non-decreasing for all x.
2. FX (0) = 0
3. limx FX (x) = 1.
Define the survival function to be the complement of the CDF, denoted SX . For all x, we write
SX (x) = 1 FX (x). Then the survival function has the following properties:
1. SX is continuous and non-increasing for all x.
2. SX (0) = 1 FX (0) = 1
3. limx SX (x) = limx [1 FX (x)] = 0.
The reason why the complement of the CDF is known as the survival function will be clear very
shortly. For exam MLC, the main focus is on random variables and the properties and descriptive
quantities of random variables. Hence, much of the basis for what is in MLC can be derived by
using principles from probabilty.
Consider the time-to-failure random variable of someone age x, denoted Tx . This random variable
should be viewed as the amount of years that someone age x will fail (or die) - hence, its domain is
in the interval (0, ). This particular random variable will have CDF FTx (t) and survival function
STx (t). To make our lives easier, we will remove the T in the subscript throughout the course;
hence, the CDF and survival functions of Tx will be Fx (t) and Sx (t) respectively.
9
10 CHAPTER 1. CONTINUOUS SURVIVAL MODELS
Notice that Fx (t) = P (Tx t). This can be interpreted as the probability that someone currently
age x will die within t years. The survival function, Sx (t) = P (Tx > t). This is the probability that
someone age x will die at some point after t years, or in other words, survive at least t years. Hence
this is the reason why the complement of the CDF is known as the survival function.
Let us now consider the intution behind the properties of the survival function using the random
variable Tx :
1. Sx (0) = 1: the probability of someone age x surviving to time 0 is 1.
2. Sx (t) is continuous and non-increasing for all t: the probability of someone age x surviving
t years decreases as t increases.
3. limt Sx (t) = 0: the probability of someone age x surviving years is 0. Eventually, all
lives age x must die.
Of course, as you should remember from probability, the probability density function (PDF) of a
continuous random variable X, denoted fX , is the derivative of the CDF of X. Similarly, for the
random variable Tx ,
d
[FTx (t)] = fTx (t) = fx (t) (1.2)
dt
which is the PDF of the time-to-failure of someone age x. Again, we drop the T in the subscript
for the PDF. We can also write the PDF in terms of the survival function:
d d
[Sx (t)] = [1 Fx (t)] = fx (t)
dt dt
d
[Sx (t)] = fx (t). (1.3)
dt
Hence, the negative derivative of the survival function of Tx is the PDF of Tx . This relationship
holds for any continuous random variable in general.
We can also integrate the PDF to get the CDF of the survival function:
t t
d
du [Fx (u)] du = fx (u) du
0 0
t
1 Fx (t) = 1 fx (u) du
0
t
Because Tx takes on values in the interval (0, ), fx (u) du = 1. Hence, for any value t such that
0
0 < t < ,
t
fx (u) du + fx (u) du = 1
0 t
t
For the rest of this course, we will use the notation (x) to refer to a person or thing whose age
is x. For exam MLC, one particular value of interest for x is the case when x = 0. Our goal is to
analyze how we can relate our general formulas mentioned so far in this chapter to those of when
x = 0.
Consider T0 , the time-to-failure random variable of someone age 0, or that of a newborn. Given
T0 > x, T0 is defined in the equation T0 = x + Tx . To understand the intuition behind this formula,
suppose that (0) is known to fail at age d. Then of course, T0 = d. After one year, (0) is now (1).
Thus, T1 = d 1 = T0 1. After another year, (1) is now (2) and T2 = (T0 1) 1 = T0 2. Repeat
this recursion for any positive integral (or integer) n and you get the relationship Tn = T0 n,
given T0 > n. It may be more intuitive to understand this relationship when written in the form
Tx = T0 xT0 > x.
Let x 0 be constant and t 0 be variable. We assume that
This assumption is definitely reasonable. In words, what (1.6) states is that the probability of (x)
surviving t years is the probability that a newborn, (0), dies within x + t years, given survival to
x years. By use of conditional probability (from exam P/1),
P (x < T0 x + t) F0 (x + t) F0 (x)
Fx (t) = P (Tx t) = = . (1.8)
P (T0 > x) 1 F0 (x)
1
In addition, we assume that Sx (t) is differentiable for all t > 0. This guarantees the existence of the force of
mortality, which we will mention later in this chapter.
12 CHAPTER 1. CONTINUOUS SURVIVAL MODELS
F0 (x + t) F0 (x)
Fx (t) = P (Tx t) =
1 F0 (x)
[1 S0 (x + t)] [1 S0 (x)]
=
S0 (x)
S0 (x) S0 (x + t)
=
S0 (x)
S0 (x + t)
=1 . (1.9)
S0 (x)
S0 (x + t)
Sx (t) = 1 [1 ]
S0 (x)
S0 (x + t)
= . (1.10)
S0 (x)
Equation (1.10) has a very intuitive interpretation: the probability that (x) survives t years is the
probability that (0) survives x + t years, given (0) survives x years. This is essentially the use of
conditional probability in survival functions.
We now focus on a quantity called the force of mortality for someone currently age x, denoted
x .2 We define
S0 (x + dx)
1
P (T0 x + dxT0 > x) S0 (x)
x = lim+ = lim+
dx0 dx dx0 dx
S0 (x) S0 (x + dx)
S0 (x)
= lim+
dx0 dx
1 S0 (x) S0 (x + dx)
= lim+ [ ]
dx0 S0 (x) dx
1 S0 (x) S0 (x + dx)
= lim+ [ ]
S0 (x) dx0 dx
1 S0 (x + dx) S0 (x)
= lim+ [ ]
S0 (x) dx0 dx
1 S0 (x + dx) S0 (x)
= lim+ [ ]
S0 (x) dx0 (x + dx) x
1 d
= [S0 (x)]. (1.11)
S0 (x) dx
2
This is an unfortunate source of notational ambiguity with X , which in exam P/1, is often used to signify the
expected value of the random variable X. For the rest of this course, assume x is used to represent the force of
mortality for someone currently age x.
1.1. CDFS, SURVIVAL FUNCTIONS, x+t 13
1 d
x = [S0 (x)]
S0 (x) dx
d
[S0 (x)]
= dx
S0 (x)
f0 (x)
= (1.12)
S0 (x)
using equation (1.3). Notice that f0 (x) = F0 (x). x can be interpreted as the instantaneous rate
of failure at age x, given survival to age x. Similarly, we can define a force of mortaility for (x) t
years after, denoted x+t . Let x 0 be fixed and t 0 be variable. x+t is defined as:
1 d
x+t = [S0 (x + t)]
S0 (x + t) d(x + t)
1 d
= [S0 (x + t)]
S0 (x + t) dt
1 d
= [S0 (x)Sx (t)] by equation (1.10)
S0 (x + t) dt
S0 (x) d
= [Sx (t)]
S0 (x + t) dt
1 d
= [Sx (t)] by equation (1.10) (1.13)
Sx (t) dt
fx (t)
= by equation (1.3) (1.14)
Sx (t)
Some references may also use x (t) to emphasize that the force of mortality depends on t. On
exam MLC, both of these notations will be used.3
Furthermore, notice that we can also write
d
[Sx (t)] d
x+t = dt = [ln Sx (t)]. (1.15)
Sx (t) dt
Thus,
t t
d
x+s ds = ds [ln Sx (s)] ds
0 0
= [ln Sx (t) ln Sx (0)]
= [ln Sx (t)] since Sx (0) = 1 ln Sx (0) = 0.
3
See Notation and Terminology used on Exam MLC, found in http://www.soa.org/Files/Edu/edu-2013-spring-
mlc-notation.pdf for the spring 2013 sitting. For the most current document, check the most current syllabus.
14 CHAPTER 1. CONTINUOUS SURVIVAL MODELS
ln Sx (t) = x+s ds
0
t
Sx (t) = exp x+s ds . (1.16)
0
where exp(x) = ex . Be sure to practice these concepts using the SOA sample question list in section
1.6 after reading the rest of this chapter.
1.2. PARAMETRIC SURVIVAL MODELS 15
1
fx (t) = (1.17)
x
t
Fx (t) = (1.18)
x
xt
Sx (t) = (1.19)
x
1
x+t = (1.20)
xt
x
E[Tx ] = (1.21)
2
( x)2
V [Tx ] = (1.22)
12
where E[Tx ] and V [Tx ] are the expected value and variance of Tx respectively,5 and t < x. In
exam MLC, the uniform distribution can referred to as de Moivres Law.
fx (t) = et (1.23)
Fx (t) = 1 et (1.24)
Sx (t) = et (1.25)
1
E[Tx ] = (1.26)
1
V [Tx ] = 2 . (1.27)
There is a valid reason why is used as the parameter for the exponential distribution. Notice that
et
x+t = t = . Thus, the force of mortality is constant for the exponential distribution. Hence,
e
the exponential distribution is sometimes referred to as the constant force distribution.
4
The symbol is used to mean in or is a member of.
5
The expected value and variance of Tx will be further discussed in the next section.
16 CHAPTER 1. CONTINUOUS SURVIVAL MODELS
Bcx (ct 1)
Sx (t) = exp [ ]. (1.29)
ln c
The proof is left to the reader. You can use equation (1.14) to derive the PDF.
In particular, the mean of X is the first moment, i.e., the case when k = 1. The variance (or the
second central moment) of X, denoted V [X] or Var[X], is defined by
Assume that X takes on nonnegative values. If the mean of X exists and is finite, an alternate
formula for the mean of X, where X (a, ) is
6
Here is how this formula is proven: suppose X is a nonnegative random variable taking values in (a, b). Then,
by definition,
b
E[X] = xfX (x) dx.
a
Using integration by parts with u = x and dv = fX (x)dx, we have du = dx and v = FX (x) + C, where C is an
arbitrary constant. Since it does not matter what value C takes, we set C = 1. Then
b
E[X] = xfX (x) dx
a
b
= b[FX (b) 1] a[FX (a) 1] [FX (x) 1] dx.
a
b
E[X] = a(1) [FX (x) 1] dx
a
b
= a + [1 FX (x)] dx. (1.35)
a
Equations (1.35) and (1.36) also apply for the case in which b , provided that E[X] exists and is finite. Even
more generally, we can show that for integral k > 0, if X (a, b), where a 0,
b
E[X ] = xk fX (x) dx
k
a
b
= b [FX (b) 1] a [FX (a) 1] [FX (x) 1]kxk1 dx
k k
a
b
= ak + k xk1 [1 Fx (x)] dx (1.37)
a
provided that E[Tx ] exists and is finite. Equation (1.39) will be very important throughout this
course.
Although not mentioned in any of the sources for exam MLC, by equation (1.37), it can be shown
that
provided that these moments exist and are finite. Equation (1.40) can be very useful in calculating
the variance and the moments of Tx .
t px : the probability that (x) survives t years. Notice that this is equivalent to Sx (t).
t px
Age x x+t
Recall that t px = Sx (t) = P (Tx > t). The arrow in the timeline above indicates where failure
or death of (x) would occur in the computation of the probability t px .
t qx : the probability that (x) dies or fails within t years. Notice that this is equivalent to
Fx (t).
t qx
Age x x+t
Recall that t qx = Fx (t) = P (Tx t). The arrow in the timeline above indicates where failure
or death of (x) would occur in the computation of the probability t qx .
7
For further details, see the article The International Actuarial Notation by Frank P. Di Paolo in the newsletter
The Actuary, March 1976 - Vol. 10, No. 3. See also International Actuarial Notation by F.S. Perryman from
Proceedings of the Casualty Actuarial Society, Vol. XXXVI, pp. 123-131, for an actuarial symbol reference.
1.4. INTERNATIONAL ACTUARIAL NOTATION: Tx 19
ut qx : the probability that (x) survives u years and dies or fails within the t years following
the u years of survival. In other words, this is the probability that (x) survives u years and
dies or fails within u + t years. This is known as a deferred mortaility probability, i.e., it
is the probability that death occurs in some interval following a deferred period. We define
ut qx = P (u < Tx u + t) = Sx (u) Sx (u + t) = Fx (u + t) Fx (u). Consider the timeline below.
ut qx
The arrow in the timeline above indicates where failure or death would occur for (x) in
computing the probability ut qx . Now consider the following timeline:
Sx (u + t)
Sx (u)
The arrows in the timeline indicate where failure or death would occur in the computations of
the probabilities Sx (u + t) = P (Tx > u + t) and Sx (u) = P (Tx > u). Think of Sx (u) Sx (u + t)
as subtracting intervals of failure, indicated by the arrows. If you subtract the interval of
failure described by Sx (u) from the interval of failure described by Sx (u + t), you get the
interval of failure for ut qx . By removing the dashed lines in the timeline below, you get the
timeline of the interval of failure for ut qx . Hence, ut qx = Sx (u) Sx (u + t).
Sx (u + t)
Sx (u)
t px = 1 t qx (1.41)
ut qx = u px u+t px (1.42)
d
[t px ] d
x+t = dt = [ln t px ] by equations (1.13), (1.15) (1.43)
t px dt
fx (t) = t px x+t by equation (1.14) (1.44)
t
t px = exp x+s ds by equation (1.16) (1.45)
0
t
It is a strongly suggested exercise for the reader to verify these relationships using their survival
function definitions. These relationships will be important throughout the course.
Another notational convention that is used is to write
ex = E[Tx ].
ex is known as the complete expectation of life.8 By the discussion in section 1.3, assuming
ex = tfx (t) dt
0
= t px dt. (1.50)
0
Intuitively, equation (1.50) makes sense. If you sum up all of the infinitesimally small intervals
that (x) can survive, you should get the expected lifetime of (x). Using equation (1.40), the kth
moment of Tx is given by, assuming each moment exists and is finite,
Needless to say, the amount of new notation is massive in this section. In section 1.7, we provide
a formula summary sheet.
When trying to find probabilities of failure given probabilities over disjoint periods, it is usually
best to use equation (1.41) to switch to survival probabilities, and then use equation (1.53) as
necessary.
Using equations (1.42) and (1.53), we can write:
ut qx = u px u+t px
= u px u px t px+u
= u px (1 t px+u )
= u px t qx+u . (1.54)
Intuitively, (1.54) makes sense. The probability that (x) fails within the age interval (x+u, x+u+t]9
is the probability that (x) survives to age x + u and fails within the following t years.
9
See the discussion on page 17. For Tx , it does not matter whether we use (), [], (], or [), since Tx is continuous.
However, for the discrete time-to-failure random variable, the meaning of these intervals does differ. A discussion
on the discrete time-to-failure random variable is in the next chapter.
22 CHAPTER 1. CONTINUOUS SURVIVAL MODELS
10
At the time of this writing, they can be found at http://www.soa.org/Files/Edu/edu-2013-spring-mlc-que.pdf.
11
At the time of this writing, the ILT can be found at http://www.soa.org/Files/Edu/edu-2013-mlc-tables.pdf.
1.7. FORMULA SUMMARY SHEET 23
fx (t) = et
Fx (t) = 1 et
Sx (t) = et
1
E[Tx ] =
1
V [Tx ] = 2 .
Gompertz Distribution:
x+t = Bcx+t
Bcx (ct 1)
Sx (t) = exp [ ]
ln c
where B (0, 1) and c > 1.
Makeham Distribution:
x+t = A + Bcx+t
Bcx
Sx (t) = exp [ (1 ct ) At]
ln c
where B > 0, c > 1, and A > B.
24 CHAPTER 1. CONTINUOUS SURVIVAL MODELS
t px = Sx (t) = 1 t qx
t+u px = t px u px+t
t qx = Fx (t) = 1 t px
ut qx = u px u+t px
= u px t qx+u
d
[ t px ] d
x+t = dt = [ln t px ]
t px dt
fx (t) = t px x+t
t
t px = exp x+s ds
0
t
t qx = t px x+t dt
0
t px = t px x+t dt
t
x+t p0
t px =
x p0
ex = E[Tx ] = tt px x+t dt
0
= t px dt
0
E[(Tx ) ] = k tk1 t px dt
k
0
Chapter 2
Before we discuss discrete survival models, we discuss symbols that will be used throughout the
rest of this text.2
2. The set of natural numbers is denoted N, which is the set {1, 2, 3, }.3 N is the set of all
whole numbers greater than or equal to 1. We say that x is an natural number by writing
x N.
3. The set of rational numbers is denoted Q, which is the set of numbers that can be written
a
in the form , where a, b Z and b 0. In other words, the rational numbers is the set of
b
all ratios of integers with a nonzero denominator. We say that x is an rational number by
writing x Q.
4. The set of irrational numbers is denoted R Q,4 which is the set of all numbers that are
not rational numbers. We say that x is an irrational number by writing x R Q.
5. The set of real numbers, denoted R, is the set R = Q (R Q). We say that x is an real
number by writing x R.
1
These are provided in the current syllabus. For the spring 2013 sitting, see http://www.soa.org/files/edu/edu-
2012-spring-mlc-studynotes.pdf.
2
Readers familiar with undergraduate real analysis may skip section 2.1 without loss of continuity. For the
reader familiar with undergraduate linear algebra or abstract algebra, it is suggested that he or she goes through
the definitions of the floor and ceiling functions.
3
Some textbooks use 0 as the lowest number of N, rather than 1.
4
The symbol is used to denote difference among sets. For example, if X, Y are sets, X Y is the set consisting
of elements that are in X but not Y . Some texts may also use to denote difference among sets.
25
26 CHAPTER 2. DISCRETE SURVIVAL MODELS
1. 5.4 = 5
2. 6.97 = 6
3. 7 = 7.
1. 5.4 = 6
2. 6.97 = 7
3. 7 = 7.
Kx = Tx . (2.1)
Kx 6 is known as the curtate time-to-failure random variable. It is the integer part of the
time-to-failure random variable Tx defined in chapter 1.7 We define another random variable as
the fractional part of Tx , denoted Rx , such that
Tx = Kx + Rx . (2.2)
Rx is a continuous random variable where Rx [0, 1). We define another random variable
Kx = Kx + 1.8 (2.3)
Kx is known as the time interval of failure random variable. For example, suppose that (85)
fails at age 87.5. Below is a timeline outlining the relationships between K85 , K85 , and T85 .
5
max{k Zk x} is the largest integer k such that k is less than or equal to x.
6
K(x) may also be used on exam MLC.
7
Notice that Kx is a discrete random variable.
8
Notice that Kx is a discrete random variable.
2.3. PROBABILITIES AND EXPECTATION OF Kx 27
R85 = 0.5
Age 85 86 87 87.5 88
T85 = 2.5
The reason why Kx is known as the time interval of failure of (x) is because it is the (annual)
interval at which (x) fails. Using the example in the timeline above, there are three age intervals:
85 to 86, 86 to 87, and 87 to 88. Since (85) fails at 87.5, (85) fails in the third age interval, i.e.,
87 to 88, and thus, K85 = 3. Also, if 1 T85 < 2, i.e., if failure were to occur between ages 86 and
87, then K85 = 2, which is the second age interval. Similarly, if 0 T85 < 1, i.e., if failure were to
occur between ages 85 and 86, then K85 = 1, which is the first age interval.
Since (85) fails at 87.5, note that K85 = 2.5 = 2.9 The fractional part of 87.5 is 0.5 = R85 . Of
course, if T85 = 2.5, K85 + R85 = 2 + 0.5 = 2.5 by the definition provided in equation (2.2).
P (Kx = k) = P (k Tx < k + 1)
= k qx by the definition on p. 1712 (2.4)
= k px qx+k by equation (1.54). (2.5)
ex = E[Kx ]. (2.6)
9
This is the rounded down value of 2.5 to the nearest integer.
10
N {0} is the set of natural numbers in addition to the number 0.
11
Not including age x + k + 1, since the floor of k + 1 is k + 1 and not k.
12
Since Tx is a continuous random variable, P (k Tx < k + 1) = P (k < Tx k + 1), which is what is stated in the
definition on p. 17.
13
Note the differences between the curtate and complete expectations of life.
28 CHAPTER 2. DISCRETE SURVIVAL MODELS
Equation (2.8) follows the same line of intuition as equation (1.50) does. The only difference
between these two equations is that (2.8) applies to the discrete case, whereas (1.50) applies to
the continuous case.
Let r N. Provided each moment of Kx exists and is finite, the rth moment of Kx is given by
E[(Kx )r ] = k r (k px k+1 px )
k=0
= 0 + (px 2 px ) + 2r (2 px 3 px ) + 3r (3 px 4 px ) +
= px (1r 0r ) + 2 px (2r 1r ) + 3 px (3r 2r ) +
= k px [k r (k 1)r ] . (2.9)
k=1
Formula (2.9) is not as easy to use as formula (1.51), so we present the formula for the second
moment. By (2.9), provided that the second moment exists and is finite,
E[(Kx )2 ] = k px [k 2 (k 1)2 ]
k=1
= k px [k 2 (k 2 2k + 1)]
k=1
= k px [2k 1]
k=1
= 2 k k px k px
k=1 k=1)
= 2 k k px e x . (2.10)
k=1
2.4. TERM EXPECTATION OF LIFE 29
E[W1 ] = E[Tx Tx n]P (Tx n) + E[nTx > n]P (Tx > n).
Thus,
n
t px x+t
E[W1 ] = t ( ) dt n qx + n n px
n qx
0
n
= tt px x+t dt + n n px
0
n
d
= t ( [t px ]) dt + n n px by equation (1.43)
dt
0
n
d
= t( [t px ]) dt + n n px . (2.11)
dt
0
d
Using integration by parts with u = t and dv = [t px ] dt,
dt
n
E[W1 ] = n n px 0 t px dt + n n px
0
n
= n n px + t px dt + n n px
0
n
= t px dt. (2.12)
0
The standard symbol for equation (2.12) is exn , which is known as the term expectation of
16
life. The term expectation of life is the expected future lifetime for (x) over the next n years
only.
14
Note that there is no symbol or standard letter given to this variable in the texts for exam MLC. We are calling
this W1 for convenience.
15
The double expectation formula for two random variables X and Y is E[X] = EY [EX [XY ]].
16
It is also known as the partial expectation of life.
30 CHAPTER 2. DISCRETE SURVIVAL MODELS
As W2 is a function of Kx , we write
n
E[W2 ] = k (k px k+1 px ) + n (k px k+1 px ) by equations (2.4) and (1.42).
k=0 k=n+1
Notice that
(k px k+1 px ) = n+1 px n+2 px + n+2 px n+3 px + n+3 px +
k=n+1
= n+1 px .
Thus,
n (k px k+1 px ) = n n+1 px . (2.13)
k=n+1
Now,
n
k (k px k+1 px ) = 0 + (px 2 px ) + 2(2 px 3 px ) + 3(3 px 4 px ) + + n(n px n+1 px )
k=0
= px + 2 px + 3 px + + n px n n+1 px
n
= k px n n+1 px . (2.14)
k=1
Therefore,
n
E[W2 ] = k px n n+1 px + n n+1 px
k=1
n
= k px . (2.15)
k=1
The standard symbol for equation (2.15) is exn , which is known as the temporary curtate
expectation of life. The temporary curtate expectation of life is the expected number of whole
years that (x) lives over the age interval (x, x + n].
2.6. SOA RELEASED QUESTIONS 31
n m n
exn = t px dt = t px dt + t px dt
0 0 m
n
=
exm + t px dt.
m
n nm
exm + t px dt =
exm + m+u px du
m mm
nm
=
exm + m+u px du
0
nm
=
exm + m px u px+m du
0
nm
exm + m px
= u px+m du
0
=
exm + m px
ex+mnm . (2.16)
(2.16) should be memorized. It comes up in quite a few of the sample questions as an extremely
convenient shortcut. We can also derive a similar formula for the discrete case. Consider for (x),
exn . Then for integral m < n,
n
exn = k px
k=1
m n
= k px + k px
k=1 k=m+1
n
= exm + k px
k=m+1
17
I thank Jim Daniel for suggesting that I add this section.
18
Credit goes to Steve Paris for giving me this step.
32 CHAPTER 2. DISCRETE SURVIVAL MODELS
Set u = k m. Then
n nm
exm + k px = exm + m+u px
k=m+1 u+m=m+1
nm
= exm + m+u px
u=1
nm
= exm + m px u px+m
u=1
nm
= exm + m px u px+m
u=1
= exm + m px ex+mnm . (2.17)
Notice that (2.17) and (2.16) are almost identical, with the exception of the on top of the e
symbol for (2.16).
(2.16) and (2.17) should be considered first priority for this exam, as they do come up in the sample
questions.
The following equation has not come up in the released questions, but may be useful. Consider
now, for (x),
ex . For n < , we have
n
ex = t px dt = t px dt + t px dt
0 0 n
=
exn + t px dt
n
w
=
exn + lim t px dt.
w
n
exn + lim t px dt =
exn + lim u+n px du
w w
n nn
wn
exn + lim
= n px u px+n du
w
0
wn
exn + n px lim
= u px+n du
w
0
=
exn + n px u px+n du
0
=
exn + n px
ex+n . (2.18)
2.6. SOA RELEASED QUESTIONS 33
Set u = k n. Then
w wn
exn + lim k px = exn + lim n+u px
w w
k=n+1 n+u=n+1
wn
= exn + lim n px u px+n
w
u=1
= exn + n px ex+n . (2.19)
Again, notice how (2.18) and (2.19) are nearly identical. The case in which n = 1 of (2.19) does
come up in the sample questions: note that this states that
Kx = Tx
Tx = Kx + Rx
Kx = Kx + 1
P (Kx = k) = P (k Tx < k + 1)
= k qx
= k px qx+k
E[Kx ] = ex
= k P (Kx = k)
k=0
= k px
k=1
E[(Kx )r ] = k px [k r (k 1)r ]
k=1
n
exn = E[min(Tx , n)] = t px dt
0
=
exm + m px
ex+mnm , for m < n
n
exn = E[min(Kx , n)] = k px
k=1
= exm + m px ex+mnm , for m < n
ex =
exn + n px
ex+n
ex = exn + n px ex+n
= px (1 + ex+1 )
Chapter 3
Life Tables
References: AMLCR, Chapters 3.2 - 3.3, 3.6 - 3.91 ; MQR, Chapters 5.4, 6.1 - 6.4, 6.6.1 - 6.6.2,
6.7.
x lx
50 500
51 490
52 470
53 430
54 370
55 290
As usual, x represents the age of a person or thing. lx represents the expected number of lives
of a population who are alive of age x.2 A life table is a table which has (at the very least)
the following information:
1. n ages x0 , x1 , , xn1 where each xi N {0} i N3 with a starting age
x(0) = min{x0 , x1 , , xn1 }4 and limiting age = max{x0 , x1 , , xn1 }, so that x(0) xi
xi ;
2. n values lx0 , lx1 , , lxn1 corresponding to each respective age xi .
The number of lives with the starting age, i.e., lx(0) , is known as the radix. Of course, for these
definitions to make sense, we assume that each xi 0 and lxi 0 xi .5 For the life table illustrated
above, x(0) = 50 is the starting age, l50 = 500 is the radix, and = 55 is the limiting age.
1
The supplementary notes to AMLCR provide background information for the sections from AMLCR that we
have omitted in this chapter (3.4 through 3.5). We will not focus too much on these sections, as they only provide
background information, and nothing really with assistance in solving exam MLC problems, judging by the released
exams. We will use the background from these chapters to give a context to the concepts in this chapter, but we
will not dedicate a whole section to them, as AMLCR does.
2
MQR uses `x instead of lx . MQRs notation will not be used on exam MLC.
3
is read as for all or for each.
4
Note that AMLCR uses x0 for the starting age. We use x(0) to indicate that it is an order statistic, i.e., the
smallest value.
5
A more rigorous approach to the life table can be found in [3, pp. 41-42].
35
36 CHAPTER 3. LIFE TABLES
Make sure you try the examples above on your own until you are comfortable calculating proba-
bilities with life tables.
We define for 0 t x(0) ,
lx(0) +t = lx(0) t px(0) . (3.1)
This makes sense, as it says that the expected number of lives who are alive with age x(0) + t is
the number who are alive at x(0) multiplied by the probability that they survive t years.
6
In this case, using (1.53) was not necessary, but depending on the information you are given in a question, you
may have to use (1.53). It will be especially useful when estimating probabilities on life tables for non-integral ages
of failure.
3.1. CALCULATING PROBABILITIES 37
Therefore,
lx+t
t px = . (3.3)
lx
We used this equation in the examples in this section without mathematically formalizing it. See
examples 1 and 3.
By equation (1.41), we have that
lx+t lx lx+t
t qx =1 = . (3.4)
lx lx
Consider the numerator lx lx+t . What is this actually describing? If we look at the example life
table in this section, for example, l50 l52 = 30. What does this quantity describe? It is the number
of deaths that have occured in the age interval (50, 52). The number of deaths over this period is
described by the symbol 2 d50 , i.e., from the ages of 50 to 52. In general, the number of deaths in
the interval (x, x + t) is t dx = lx lx+t .7
Thus,
t dx
t qx = . (3.5)
lx
We used this equation in example 2 without mathematically formalizing it. We could also present
a formula for ut qx , but as you see from example 4, it is best just to use equation (1.54) with (3.3)
and (3.4).
7
AMLCR does not mention this notation, nor is it in International Actuarial Notation by Perryman. However,
both of these sources do mention dx , the number of deaths that occur in the age interval (x, x + 1). The Notation
and Terminology used on Exam MLC document for spring 2013 does not provide any comment on whether or not
this notation will be used.
38 CHAPTER 3. LIFE TABLES
As an exercise, complete the following life table. The values that have been filled in are the ones
stated in the example, in addition to d55 . Round final answers to three decimal places at most.
x lx px qx dx
50 500 0.98 0.02
51 490
52 470
53 430
54 370
55 290 100
The solution is given below. Due to rounding, p51 + q51 does not equal 1.
x lx px qx dx
50 500 0.98 0.02 10
51 490 0.96 0.041 20
52 470 0.915 0.085 40
53 430 0.86 0.14 60
54 370 0.784 0.216 80
55 290 0.655 0.345 100
Here are some observations to notice about life tables that we can get from this particular life
table:
8
[1, pp. 102-103] covers the hyperbolic/Balducci assumption; however, this has since been dropped from the
MLC syllabus and will not be covered in this text.
3.2. FRACTIONAL AGE ASSUMPTIONS 39
Let x be the independent variable and lx be the dependent variable. From Calculus, the slope, m,
of the secant line connecting the points (x, lx ) and (x + 1, lx+1 ) is
lx+1 lx
m= = lx+1 lx .9 (3.6)
(x + 1) x
We assume this slope m is constant throughout the interval (x, x + 1). This would imply that
lx+s = (lx+1 lx )(x + s) + b for some b R.10 Of course, lx+0 = lx , so that lx = (lx+1 lx )x + b. Thus,
we can write
Now here is where things get very interesting: define x and s as stated earlier in this section. Then
lx lx+s
s qx =
lx
lx (1 s)lx slx+1
= by equation (3.7)
lx
slx slx+1
=
lx
lx lx+1
= s( )
lx
= sqx by equation (3.4). (3.8)
Equation (3.8) is known as the uniform distribution of deaths (UDD) assumption.11 The
reason why this is known as the uniform distribution of deaths assumption is the following: recall
that Tx = Kx + Rx by equation (2.2). It can be shown that (3.8) is equivalent to stating that, for
x N {0}, Rx follows a uniform distribution in (0, 1), and Kx and Rx are independent.12
Of course, you can use previously developed relationships among formulas to gather many new
ways to rewrite formulas under the UDD assumption, but we will focus on only two of these in
particular.
Notice that, for variable s (0, 1), by equation (3.8),
d
[s qx ] = qx . (3.9)
ds
d d d
x+s = [ln s px ] = [ln((px )s )] = [s ln px ] = ln px . (3.15)
ds ds ds
Notice that x+s does not depend on s, hence, it is constant in the age interval (x, x + 1), so we
write x = x+s . This is why the exponential assumption of lx+s is called constant force (CF)
assumption. This leads to px = exp (x ), or (px )s = s px = exp (sx ). Equivalently, by equation
(1.25), we can say that Tx follows an exponential distribution in the age interval (x, x + 1).
3.3. SELECT AND ULTIMATE LIFE TABLES 41
Valuation of Insurances
References: AMLCR, Chapter 4; MQR, Chapters 7.1 - 7.5, 10.1.
This is the chapter where AMLCR and MQR start presenting very different perspectives on the
same material. We will do our best to include both perspectives in the discussion in this text.
4.1 Introduction
In this section, we focus on the valuation of insurances. An insurance makes a single payment
due to the occurrence of a defined random event. The main focus of this course is on life insurance,
which makes a single payment due to death of a person. However, the concepts presented here can
easily be extended to insurances for machinery, cars, homes, etc.
From exam FM/2, one possible way to value an insurance is to take the present value of the cash
flow that the company will pay upon the persons death. But there is one problem: how do we
know when the person will die?
If (x) possesses a $5,000 insurance which pays right at the moment of his or her death, which, say,
were to occur 25 years from now, then from FM/2, we know that the present value of the insurance
would be 5000v 25 .1 But the problem is, if we do not currently know when the death of (x) would
occur, all we would know is that if (x) were to die Tx years from now, the present value of the
insurance would be 5000v Tx . Notice that Tx , the time-to-failure random variable, is brought into
the present value of the insurance.
So, given the present value of the insurance 5000v Tx , it seems that it would be logical to value
this insurance by taking the expected present value (EPV) of this insurance, i.e., E[5000v Tx ].
Another name for the EPV is the actuarial present value (APV).2 Of course, in order to value
this insurance, we would need to make assumptions about the distribution of Tx . This discussion
forms the basis for pricing continuous insurances.
1 1
Where v = , i being the effective annual interest rate.
2
1 +i
AMLCR also calls the APV/EPV the Actuarial Value, and MQR also calls the APV/EPV the Net Single
Premium (NSP), but none of these terms will be in exam MLC.
45
46 CHAPTER 4. VALUATION OF INSURANCES
Now another thing to consider is the following: realistically, insurances do not pay right at the
moment of death of a person. If we consider the discrete-time-to-failure random variables Kx and
Kx , we could find an expression for the present value of an insurance which uses one of these two
random variables. Suppose (x) were to die Tx = 24.5 years from now. When would it be most
logical for the insurance of $5,000 to pay?
What about the beginning of the year of death? Paying at Tx = 24 = Kx years from now does
not make any sense, as there would be no reason for the insurance company to pay, as death has
not occurred yet. Thus, it would be most logical for the company to pay the insurance benefit
at the end of the year of death, i.e., Kx + 1 = 24 + 1 = 25 = Kx years from now. So the expression
for the present value of such an insurance would be 5000v Kx = 5000v Kx +1 , and the APV/EPV of
such an insurance would be E[5000v Kx ] = E[5000v Kx +1 ], and of course, we would need to make
assumptions about the distributions of Kx and Kx in order to price this insurance. This discussion
forms the basis for pricing discrete insurances.
Where AMLCR and MQR differ is in their choices of random variables to price the discrete
insurances: AMLCR uses Kx , whereas MQR uses Kx . Another way in which these two texts differ
is their notation of the present value random variable for the different types of insurances presented
in this chapter. AMLCR and the SOAs released questions use Z to represent the present value
(PV) random variable for an insurance, regardless of the insurance type. MQR decorates Z using
symbols which are similar to those from exam FM/2, depending on the insurance type. All three
sources, however, have the same symbols for the APV/EPV for each type of insurance.
We will go into further detail on the symbols that are used as we go throughout this chapter.
Most of the time, we will be using the MQR notation for the present value random variable for the
different types of insurances. Although this system of notation is not used by the SOA or AMLCR,
this system of notation makes it easier to relate each type of insurance with its respective APV
symbol.
47
48
Table of Continuous Insurance Types(fx (t) = t px x+t )
PV Random Name of Payment Description Definition Expected Value
Variable Insurance
Zx Whole Life Pays 1 at the moment Zx = v Tx Ax = v t fx (t) dt
0
Insurance of death
v Tx , Tx n
n
Zxn
1
Term Insurance Pays 1 at the moment 1
Zxn = A1xn = v t fx (t) dt
0, Tx > n
0
of death if death
occurs within n years
0,
Tx n = v t fx (t) dt
n Zx Deferred Whole Pays 1 at the moment n Zx = T n Ax
v , T x > n
x
n
Life Insurance of death if death
occurs after n years
u+n
1 1 0,
Tx u or Tx > u + n 1
u Z xn Deferred Term Pays 1 at the moment Z
u xn = T u Axn = v t fx (t) dt
v , u < Tx u + n
x
u
Insurance of death after the
u-year deferral period
We now go into detail on moments of the present value random variables of insurances.
2
Ax = e2k P (Kx = k). (4.1)
k=1
n
2
A1xn = e2k P (Kx = k) (4.7)
k=1
1 2 1
V [Zxn ]= Axn (A1xn )2 (4.8)
1
V [bZxn ] = b2 [2 A1xn (A1xn )2 ]. (4.9)
n
2
A1xn = e2t fx (t) dt (4.10)
0
V [Zxn
1
]= 2 1
Axn (A1xn )2 (4.11)
V [bZ 1 ] = xn b2 [2 A1xn (A1xn )2 ]. (4.12)
2k
2
n Ax = e P (Kx = k) (4.13)
k=n+1
2 2
V [n Zx ] = n Ax (n Ax ) (4.14)
V [b n Zx ] = b2 [2 n Ax (n Ax )2 ]. (4.15)
For n Zx , we have
2 2k
n Ax = e fx (t) dt (4.16)
n
V [n Zx ] = 2 2
n Ax (n Ax ) (4.17)
V [b n Zx ] = b2 [2 n Ax (n Ax )2 ]. (4.18)
1
For u Z xn , we have
u+n
2 1
u Axn = e2k P (Kx = k) (4.19)
k=u+1
1 2 1 1 2
V [u Z xn ]= u Axn (u Axn ) (4.20)
1
V [b u Z xn ] = b2 [2 u Axn
1 1
(u Axn )2 ]. (4.21)
4.3. VARIANCES AND MOMENTS 51
For u Z xn
1
, we have
u+n
2 1 2t
u Axn = e fx (t) dt (4.22)
u
V [u Z xn
1
] = 2 u Axn
1 1
(u Axn )2 (4.23)
V [b u Z xn
1
] = b2 [2 u Axn
1
(u Axn
1
)2 ]. (4.24)
Axn1 = e2n n px
2
(4.25)
V [Z xn1 ] = 2 Axn1 (Axn1 )2 = 2 Axn1 (n Ex )2 (4.26)
V [bZ xn1 ] = b2 [2 Axn1 (Axn1 )2 ] = b2 [2 Axn1 (n Ex )2 ]. (4.27)
1
Cov(Zxn , Z xn1 ) = E[Zxn
1
Z xn1 ] E[Zxn
1
]E[Z xn1 ]
1
= 0 E[Zxn ]E[Z xn1 ]
= A1xn n Ex . (4.29)
V [Zxn ] = V [Zxn
1
] + V [Z xn1 ] + 2 (A1xn n Ex ). (4.30)
3
Note that there is no notation for n Ex calculated with force of interest 2.
4
This is an application of the following formula for two random variables X and Y : V [X + Y ] = V [X] + V [Y ] +
2 Cov(X, Y ), where Cov(X, Y ) = E[XY ] E[X]E[Y ].
52 CHAPTER 4. VALUATION OF INSURANCES
= et et dt
a
b
= e(+)t dt
a
1
= [ ] [e(+)b e(+)a ]
+
[e(+)a e(+)b ]
= . (4.31)
+
1
If Tx follows a uniform distribution (or de Moivres law), we have fx (t) = and assuming
x
0 a < b x, similarly,
b b
t t 1
v fx (t) dt = e ( x ) dt
a a
b
1
= et dt
x
a
1 1
= ( ) (eb ea )
x
ea eb
= . (4.32)
( x)
4.4. MORTALITY ASSUMPTIONS ON THE ENTIRE DOMAIN OF Tx 53
For the uniform distribution, set a = 0 and b = n x in equation (4.32). This leads to
e(0) en 1 en
=
( x) ( x)
a
n
= . (4.36)
x
5 1 vn
This uses the fact that, from exam FM/2, a
n = .
54 CHAPTER 4. VALUATION OF INSURANCES
Ax = A1xn + n Ax . (4.37)
To prove this, use the definitions from the Table of Continuous Insurance Types. For equation
(4.32), inputting the values above for a and b does not lead to a useful simplification of n Ax ; in
fact, when simplified, it leads to (4.37). However, for the constant force distribution, simplifying
leads to
= Ax A1
n Ax xn
= Ax Ax (1 n Ex ) by equation (4.35)
= Ax n Ex . (4.38)
We could also write A1xn u Ex for equation (4.39). For equation (4.32), assuming u + n x, we
have:
eu e(u+n) eu (1 en )
=
( x) ( x)
1 en
= eu [ ]
( x)
= eu A1xn by equation (4.36). (4.40)
x (k 1) (x + k 1) 1
k1 px qx+k1 = [1 ]
x (x + k 1)
x k + 1 (x + k 1) + (x + k 1) + 1
= [ ]
x xk+1
1
= . (4.41)
x
So, interestingly enough, the PMF of Kx is the same form as the PDF of Tx under the uniform
distribution. However, for the exponential (constant force) distribution, the PMF of Kx and the
PDF of Tx are not of the same form. Under the exponential distribution, we have:
k1 px qx+k1 = e(k1) (1 e ) 6
= ek e (1 e )
= ek (e 1). (4.42)
Now we proceed to valuing the insurances. Suppose that an insurance pays a unit benefit to (x) if
(x) dies in the age interval (x + a 1, x + b) at the end of years x + a through x + b. The value of
this insurance is given by
b
v k k1 px qx+k1 . (4.43)
k=a
6
Do not forget that qx = 1 qx .
56 CHAPTER 4. VALUATION OF INSURANCES
7
The derivations on this page use the following facts:
1 xn+1
1. 1 + x + x2 + + xn = for x 1. We implicitly assume that i > 0, so that 0 < v < 1.
1x
i
2. 1 v = d = , from exam FM/2.
1+i
1 vn
3. an = , from exam FM/2.
i
4.4. MORTALITY ASSUMPTIONS ON THE ENTIRE DOMAIN OF Tx 57
(e 1)e(+)
[1 e(+)n ] = Ax [1 e(+)n ]
1 e(+)
= Ax (1 n Ex ). (4.48)
Under the uniform distribution, set a = 1 and b = n x in equation (4.45). Then we have
an
. (4.49)
x
58 CHAPTER 4. VALUATION OF INSURANCES
Ax = A1xn + n Ax (4.50)
to compute n Ax . This can be proven easily using the definitions from the Table of Discrete In-
surance Types. Similarly, for discrete insurances, it can be shown that under the exponential
distribution,
n Ax = Ax n Ex . (4.51)
1 e(+)(u+n(u+1)+1) (+)(u+1) 1 e
(+)n
(e 1)e(+)(u+1) [ ] = (e
1)e [ ]
1 e(+) 1 e(+)
(e 1)e(+)
= [1 e(+)n ] e(+)u
1 e(+)
= e(+)n A1xn . (4.52)
Similar to section 4.4.1.5, we could also write A1xn u Ex . For equation (4.45), set a = u + 1 and
b = u + n x. Then we have
v u+11 vu
( ) au+n(u+1)+1 = ( ) an
x x
= v u A1xn . (4.53)
Ax = et t px x+t dt
0
k+1
= et t px x+t dt.
k=0
k
Notice that t R and k N {0}. This is similar to saying that Tx has domain in the real numbers,
and Kx has domain in the nonnegative integers.8 So the fractional part, which we typically call Rx
but will label s in our integration, is defined by t = k + s. Using this substitution with s = t k as
the new variable of integration, we have ds = dt and
k+1 k+1k
t
e t px x+t dt = e(k+s) k+s px x+k+s ds
k=0 k=0
k kk
1
= e(k+s) k+s px x+k+s ds
k=0 0
1
= e(k+s) k px s px+k x+k+s ds
k=0 0
1
= e(k+s) k px qx+k ds by the UDD assumption
k=0 0
1
= e(k+s) k qx ds.
k=0 0
8
Note that we are not using Kx here, as Kx does not take value 0, which is in the summation.
60 CHAPTER 4. VALUATION OF INSURANCES
1
n
k i
e k 1 qx e(s1) ds = A1xn ( ) . (4.56)
k =1
0
We can also extend our results to u A1xn and get a similar result. The main idea to get from
this section is for any insurance which has a level present value random variable for one straight
i
(and possibly infinite) period, under UDD, the Continuous EPV = Discrete EPV rule applies.
Examples of insurances which do not follow this rule:
1. Endowment insurance: it does not have a level present value random variable. If failure occurs
before n years, the present value random variable takes value v Tx . If failure occurs after n
years, the present random variable takes value v n . If the present value random variable were
to take value v Tx after n years, the rule would apply (but it does not - actually, this would
i
be a whole life insurance). However, you can use the rule Axn = A1xn + n Ex = A1xn + n Ex if
given the UDD assumption.
v k , Kx n
2. Define a discrete insurance with present value random variable Z = 0, n < Kx n + 2
k
v , Kx > n + 2.
Though this insurance has a level present value random variable (equal to v k when it does
pay the benefit), the period at which it pays the benefit is broken up by the period when it
pays nothing, when n < Kx n + 2.
9
This is to be consistent with the Kx notation that we have used in this text.
4.5. UDD ASSUMPTION: CONTINUOUS-DISCRETE RELATIONSHIPS 61
i
3. Note that Axn1 Axn1 , since there is no such thing as Axn1 .
We can also extend our results to kth moments of present value random variables of insurances,
assuming the moments exist and are finite. In order to not mix up the variables, we will refer to
kth moments as pth moments in this section. Recall that by the theorem presented in section 4.3,
we can find the pth moment of an insurance by multiplying the force of interest by p.
In the case of the whole life insurance, modifying (4.54) and using the theorem presented in section
4.3,
1
pk ep
qx ep(s1) ds = epk ) (ep 1)
e k 1 k 1 qx (
k =1 k =1 p
0
ep 1
= epk
k 1 qx ( ). (4.58)
k =1 p
When p = 2, we have
e2 1 e2 1
e2k
k 1 qx ( ) = 2 Ax ( ). (4.59)
k =1 2 2
(1 + i)2 1
In terms of i in the numerator, (4.59) would be 2 Ax [ ]. Similarly, for any insurance
2
which has a level present value random variable for one straight random variable for one straight
(1 + i)2 1
(and possibly infinite) period, under UDD, the 2 Continuous EPV = 2 Discrete EPV * [ ]
2
rule applies.
62 CHAPTER 4. VALUATION OF INSURANCES
R85 = 0.5
(2) 1
Notice that K85 deals with periods rounded to the nearest of a year, or the nearest semiannual
2
(2) 1
period. As T85 = 2.5, we have K85 = 2.5 2 = 2.5 and J85 = (2)(2.5) + 1 = 6.
2
To be consistent, we will use Jx in the notation of the EPV/APV of insurances.
(m) (m)
To make our lives a little easier, let us find the probability mass function for Kx . If Kx = k,
then
(m) k k 1
P (Kx = k) = P ( Tx < + )
m m m
= k 1 qx . (4.62)
m
(m) j1
P (Jx = j) = P (Kx = )=
j1 1 qx . (4.63)
m m m
10 (m)
This random variable comes from MQR. You would think that MQR would use Kx instead of Jx to denote
this symbol, but it does not.
4.6. mTHLY INSURANCES 63
m
i(m) i(m)
In order to maintain consistent notation, recall that for m > 0, (1 + ) = 1 + i, and is the
m m
i(m) 1
effective mthly rate. We define the mthly discount factor v ( )= .11 When m = 1,
m i(m)
1+
m
1
we will use the usual v = notation.
1+i
i(m)
For m 1, we will often use k = , or v(k). The reason why we are using this notation is because
m
MQR uses v to denote the mthly discount factor, which is improper notation, as it only applies to
the effective annual interest rate.
The theorem presented in 4.3 also applies to mthly insurances. Similar to what has been presented,
(m) (m)
we have the notations 2 Ax , 2 A(m)1xn , and 2 Axn to represent the EPV/APV of their respective
insurances calculated under the force of interest 2.
11 i(m)
None of the texts for MLC use the v ( ) notation. We use it to remove ambiguity from v, which MQR uses
m
for the mthly discount factor, although v is reserved for the annual interest rate.
64
Table of mthly Insurance Types (P (Jx = j) = j1 1
qx )
m m
PV Random Name of Payment Description Definition Expected Value
Variable Insurance
(m) (m) (m)
Zx Whole Life Pays 1 at the end of Zx = [v(k)]Jx Ax = [v(k)]j P (Jx = j)
j=1
Insurance the m1 th period of
death
[v(k)]Jx , Jx n
mn
Z (m)1xn Term Insurance Pays 1 at the end of Z (m)1xn = A(m)1xn = [v(k)]j P (Jx = j)
0, Jx > n
j=1
the m1 th period of
death if death occurs
within n years
(m) (m) (m)
Zxn Endowment Pays 1 at n years or Zxn = Z (m)1xn + Z xn1 Axn = A(m)1xn + n Ex
Insurance at the end of the m1 th
period of death if
death occurs within n
years, whichever
comes first
If the insurance pays bk at the end of the year of the death of (x) (i.e., the insurance is discrete)
if death occurs in the age interval (x + a 1, x + b), where b N, then its APV/EPV is given by
b
bk v k k1 px qx+k1 . (4.65)
k=a
1
If the insurance pays bj at the end of the th of a year of the death of (x) (i.e., the insurance is
m
a1 b
mthly) if death occurs in the age interval (x + , x + ), then its APV/EPV is given by
m m
b
bj [v(k)]j j1
m
px 1 qx+ j1 .
m m
(4.66)
j=a
12
For a partial proof of this relationship, see [1, pp. 138-139].
66 CHAPTER 4. VALUATION OF INSURANCES
1. (IA)x is the APV of a discrete whole life insurance for (x) (implied by the Ax ). If (x) dies
in year 1, the benefit will pay 1 at the end of the first year; if (x) dies in year 2, the benefit
will pay 2 at the end of the second year; and so on (the annual increase is implied by the I).
In general, this particular insurance pays k at time k, where k is an natural number. Thus,
the APV is given by
(IA)x = kv k k1 px qx+k1 dt. (4.67)
k=1
x is the APV of a continuous whole life insurance for (x) (implied by the Ax ). This
2. (I A)
insurance pays at the moment of death of (x). Since we have I in here, the insurance benefit
increases annually. So if (x) dies in the first year, the benefit will pay 1 at the moment of
death; if (x) dies in the second year, the benefit will pay 2 at the end of the second year at
the moment of death, and so on. Observe that we can describe the benefit amount at time
t for (x) as 1, if (x) dies in the first year; 2, if (x) dies in the second year, and so on. In
general, the insurance pays t at every integer time t, so the APV is given by
1 2 3
x = 1v fx (t) dt + 2v fx (t) dt + 3v t fx (t) dt + .
(I A) t t
(4.68)
0 1 2
13
Note that I (m) , although theoretically possible, is not presented in any of the textbooks for MLC.
4.8. INSURANCES WITH VARIABLE BENEFITS 67
1
0 1 2 3 4
t
3. (IA)
x is the APV of a continuous whole life insurance for (x) (implied by the Ax ). The
so that at any time t, the insurance
increase is also happening continuously (implied by the I),
pays t. (Notice that the benefit amount starts off at 0 and increases continuously.) The APV
of this insurance is given by
(IA)
x = tv t fx (t) dt. (4.70)
0
The concepts here can easily be extended to insurances besides whole life insurances.
Instead of the I, we can also use D.14 Note that D can only be used for insurances which
have a finite term. D, as you may recall from exam FM/2, refers to a decrease of some amount
D, D(m) .15 The same interpretations for the different
each period. We can manipulate D like so: D,
forms of I apply to the analogous forms for D. We present only one example here; extensions can
easily be made to other types of decreasing insurances.
Example.
1. (DA) 1 is an n-year continuous term insurance given to x. If (x) dies in year 1, the benefit
xn
pays n at the moment of death; if (x) dies in year 2, the benefit pays n 1 at the moment
of death; and so on, until if (x) dies in the nth year, the benefit pays 1 at the moment of
death. In general, it can be shown that the insurance pays n t + 1 at the moment of death
at time t.
14
AMLCR does not cover decreasing insurances, interestingly enough.
15
Like with I (m) , D(m) is not presented in any of the textbooks for MLC.
68 CHAPTER 4. VALUATION OF INSURANCES
(4.73) is the recursive relationship of whole life insurances for (x). For mthly insurances, we have:
(m)
Ax = [v(k)]j j1 1 qx
m m
j=1
= v(k)1/m qx + [v(k)]2 1/m px 1/m qx+1/m + [v(k)]3 2/m px 1/m qx+2/m + (4.74)
(m)
Ax+1/m = [v(k)]j j1 1 qx+1/m
m m
j=1
= v(k)1/m qx+1/m + [v(k)]2 1/m px+1/m 1/m qx+2/m + [v(k)]3 2/m px+1/m 1/m qx+3/m + . (4.75)
We can write
(m)
Ax = v(k)1/m qx + [v(k)]2 1/m px 1/m qx+1/m + [v(k)]3 2/m px 1/m qx+2/m +
= v(k)1/m qx + v(k)1/m px {v(k)1/m qx+1/m + [v(k)]2 1/m px+1/m 1/m qx+2/m + }
(m)
= v(k)1/m qx + v(k)1/m px Ax+1/m . (4.76)
4.10. ACTUARIAL ACCUMULATED VALUE 69
Hence, we have
APV of Insurance
= AAV of Insurance at time n.17 (4.77)
E
n x
Note that (4.77) will only apply to insurances with a finite term.
Throughout the course, you will notice that n Ex will function like a discount factor for MLC,
as v did for FM/2. Using this intuition, you can derive the following formulas:
Ax = Ax+n n Ex (4.78)
Ax = Ax+n n Ex . (4.79)
16
Credit goes to Prof. Herschel Day for providing me with this intuition.
17
MQR uses t kx to represent the AAV of a term insurance with term t years, and t kx to represent the AAV of
a continuous term insurance with term t years. See [1, p. 224] and [1, p. 229]. However, none of these symbols are
mentioned in the Notation and Terminology used on Exam MLC note.
70 CHAPTER 4. VALUATION OF INSURANCES
As you go through these questions, when in doubt, use the most general formulas given in chapter
4.8.1.
4.12. FORMULA SUMMARY SHEET 71
Theorem. Following the AMLCR/SOA convention, let Z be the PV random variable of an insur-
ance in either the Table of Discrete or Continuous Insurance Types. The kth moment of the PV
random variable of any insurance with annual (level) benefit b > 0, force of interest = ln(1 + i),
and expected value E[Z] = A() is given by E[(bZ)k ] = bk A(k), given that each moment ex-
ists and is finite. Equivalently, if E[Z] = A(i), where i is the effective annual interest rate, then
E[(bZ)k ] = bk A[(1 + i)k 1].
Actuarial Present Values
Exponential (Constant Force) Distribution
Notice that none of these formulas rely on the age x. Hence, they are equivalent for all ages.
n Ex = e(+)n
[e(+)a e(+)b ]
General Continuous Equation
+
Ax =
+
Axn = Ax (1 n Ex )
1
u A
1 = u Ex A1
xn xn
1 e(+)(ba+1)
General Discrete Equation (e 1)e(+)a [ ]
1 e(+)
qx
Ax =
qx + i
1
Axn = Ax (1 n Ex )
1
u Axn = u Ex A1xn
72 CHAPTER 4. VALUATION OF INSURANCES
ea eb
General Continuous Equation
( x)
a
x
Ax =
x
a
n
A1xn =
x
1 u 1
u Axn = e Axn
v a1
General Discrete Equation ( )a
x ba+1
ax
Ax =
x
an
A1xn =
x
1 u 1
A
u xn = v Axn
General Cases
b
mthly Insurances
(m) 1
Kx = mTx
m
(m)
Jx = mKx + 1
(m)
P (Kx = k) = k 1 qx
m
P (Jx = j) =
j1 1 qx
m m
1
v(k) =
i(m)
1+
m
Recursive Relationships
Other Equations
APV of Insurance
= AAV of Insurance at time n
n Ex
1
Cov(Zxn , Z xn1 ) = A1xn n Ex
Ax = A1 + n Ax
xn
Ax = A1xn
+ n Ax
Ax = Ax+n n Ex
Ax = Ax+n n Ex
n ax n!
x e dx = an+1
0
74 CHAPTER 4. VALUATION OF INSURANCES
How do we price continuous and discrete insurances with all four of these variables non-constant?18
Suppose that at time t, we have a non-constant force of interest t . From exam FM/2, the present
value of a payment of 1 at time t is given by
t
exp r dr .
0
Suppose that for a discrete insurance, P (Kx = k) = k1 qx and for a continuous insurance, Tx has
probability density function fx (t).
Suppose now that the discrete insurance pays at times k1 , k2 , . . . , kn , with n and kn possibly being
infinite. Also, suppose T = {[t1 , t2 ], [t3 , t4 ], [t4 , t5 ], . . . , [tn1 , tn ]}, where n and tn can possibly be
infinite, is the set of intervals during which the continuous insurance could pay a benefit. Let bk
be the benefit at time k (for a discrete insurance) and bt be the benefit at time t (for a continuous
insurance).
In general, we can write
k
APVDiscrete = bk exp r dr k1 qx (4.81)
k=k1 ,k2 ,...,kn
0
t
APVContinuous = bt exp r dr fx (t) dt. (4.82)
T 0
where means to sum over all ki , i = 1, 2, . . . , n, and means to integrate over all intervals
k=k1 ,k2 ,...,kn T
in T .
Finding the actuarial accumulated value (or AAV) of a very general insurance is even more com-
plicated. As was said in section 4.10, we would divide by n Ex . However, the force of interest is not
constant. Recall that the AAV of an insurance can only be computed if the term of the insurance
is finite. Instead of dividing by n Ex , using the discrete model, we would have to divide by some
form of kn Ex . For continuous insurances, we would need to divide by some form of tn Ex .
18
Notice here - I am not including mthly insurances in this discussion. Actually, mthly insurances are just discrete
insurances paid at a different frequency.
4.13. APPENDIX: FULLY-VARIABLE INSURANCES 75
Therefore,
APVDiscrete
AAVDiscrete (kn ) = (4.85)
kn Ex
APVContinuous
AAVContinuous (tn ) = , (4.86)
tn Ex
where (kn ) and (tn ) are used to denote the times at which the AAV is valued (at times kn and tn ,
respectively).
76 CHAPTER 4. VALUATION OF INSURANCES
Chapter 5
Valuation of Annuities
1. The annuity-immediate pays at the end of each year. Below is an example of a three-year
annuity-immediate which pays 1 each year.
Payment 1 1 1
Time 0 1 2 3
a3
v2
v3
The present value of this particular annuity-immediate, assuming annual effective annual
interest rate i 0, is
1 v3
a3 = v + v 2 + v 3 = .
i
Of course, if i = 0, then the present value is simply 3.
1
Remember: insurances are single payments, whereas annuities are streams of payments.
77
78 CHAPTER 5. VALUATION OF ANNUITIES
2. The annuity-due pays at the beginning of each year. Below is an example of a three-year
annuity-due which pays 1 each year.
Payment 1 1 1
Time 0 1 2 3
a
3
v2
The present value of this particular annuity-due, assuming an annual effective annual interest
rate i 0, is
1 v3
3 = 1 + v + v 2 =
a ,
d
i
where d = . Of course, if i = 0, then the present value is simply 3.
1+i
In general, we can write for i 0,
1 vn
an = v + v 2 + + v n =
i
n1 1 vn
n = 1 + v + + v
a = .
d
For i = 0, an = a
n = n.
In general, for i 0, it can also be shown that
an1 + 1 = a
n
n 1 + v n .
an = a
The future value of the annuity-immediate is given by
(1 + i)n 1
sn = (1 + i)n an = ,
i
which is the value of an annuity paying 1 at the end of each year at the time of the last payment.
Payment 1 1 1
Time 0 1 2 3
s3
(1 + i)2
1+i
5.1. REVIEW FROM FM/2 79
(1 + i)3 1
s3 = (1 + i)2 + (1 + i) + 1 = .
i
The future value of the annuity-due is given by
(1 + i)n 1
sn = (1 + i)n a
n = ,
d
which is the value of the annuity paying 1 at the end of each year at the time after the last payment.
Payment 1 1 1
Time 0 1 2 3
(1 + i)3 s3
(1 + i)2
1+i
(1 + i)3 1
s3 = (1 + i)3 + (1 + i)2 + (1 + i) = .
d
Payment 1 1 1
Time 0 1 2 3
The present value of this annuity is found by discounting 1 at each infinitesimally small time
period, or
3
1 v3
3 = v t dt =
a ,
0
where is the force of interest. The future value of this annuity is found by multiplying a
3 by
3
(1 + i) , or
(1 + i)3 1
s3 = (1 + i)3 a
3 = .
80 CHAPTER 5. VALUATION OF ANNUITIES
and
(1 + i)n 1
sn = (1 + i)n a
n = .
For = 0, a
n = n.
(m) 1 1 vn
a
n = (1 + v 1/m + v 2/m + + v n1/m ) = (m) .
m d
1 1
If the mthly annuity pays at the end of every th of a year for a total of n years, the annuity
m m
has present value
(m) 1 1 vn
an = (v 1/m + v 2/m + + v n/m ) = (m) .
m i
5.2 Introduction
We approach pricing annuities much like how we price insurances. Insurances, as we saw in
chapter 4, make a single payment due to death of a person. However, annuities make payments
if the person is still alive at the time the payment is made.
Consider (x), whose death is to occur 25.5 years from now. Suppose (x) purchases an annuity-
immediate which pays 1 annually until his or her death. Annuities which pay only if the purchasers
are alive are known as life annuities. If (x) dies 25.5 years from now, the price of the annuity
is given by a25 , as this is the present value of 25 payments - one at the end of each year during
which (x) is still alive.
But, as with insurances, there is no way to know when (x) will die. So how do we price a life
annuity?
Similar to insurances, we will have to introduce the random variables Tx , Kx and Kx somewhere
in the pricing formulas.
5.3. WHOLE-LIFE ANNUITIES 81
Yx = a
Kx +1 = a
Kx (5.1)
Yx = aKx = aKx 1 (5.2)
=aKx +1 1 = a
Kx 1
= Yx 1 (5.3)
Yx = a
Tx (5.4)
(m) (m)
Yx = a (5.5)
Jx /m
(m) (m) 1
Yx = Yx .2 (5.6)
m
Alternatively, using the notation in section 4.6, we could write
(m) 1
Yx = a (5.7)
m Jx k
(m) (m) 1
Yx = aJx 1k = Yx , (5.8)
m
i(m)
where k indicates that the rate used to value the annuity is k = , and instead of v, v(k) is used
m
(as defined in section 4.6).
The expected values of these variables are given by
E[Yx ] = a
x (5.9)
E[Yx ] = ax (5.10)
E[Yx ] = a
x (5.11)
(m) (m)
E[Yx ] = a x (5.12)
(m) (m)
E[Yx ] = ax . (5.13)
(5.9) through (5.13) are actuarial present values (APVs) of annuities. (5.9) is the APV of a
whole-life annuity-due, (5.10) is the APV of a whole-life annuity-immediate, (5.11) is the APV of
a whole-life continuous annuity, (5.12) is the APV of a whole-life mthly annuity-due, and (5.13) is
the APV of a whole-life mthly annuity-immediate.
2
I am currently working on trying to find the formula for this random variable in terms of the mthly annuity-
immediate. MQR does write this formula but in confusing notation. AMLCR does not present this formula.
82 CHAPTER 5. VALUATION OF ANNUITIES
ax = ax 1 (5.17)
V [Yx ] = V [Yx ]. (5.18)
1 v Kx 1
1 (1 + i)Ax
E[Yx ] = ax = (5.20)
i
(1 + i)2 Ax
V [Yx ] = . (5.21)
i2
I suggest that if you are studying for exam MLC, do not memorize (5.19) through (5.21) and
memorize the relationships given by (5.17) and (5.18) to get the same values as (5.20) and (5.21).
1 v Tx
Since Yx = a
Tx = , since Zx = v Tx for all Tx (0, ),
1 Zx
Yx = . (5.22)
5.5. NON-INSURANCE ANNUITY APV FORMULAS 83
1 Ax
E[Yx ] = a
x = (5.23)
2Ax (Ax )2
V [Yx ] = . (5.24)
2
(m) 1 1 1 [v(k)]Jx
As Yx = a Jx k = ( ), we can write
m m d(k)
(m)
(m) 1 1 Zx
Yx =
m d(k)
(m)
(m) 1 1 Ax
a
x =
m d(k)
2A (m) (m)
(m) x (Ax )2
V [Yx ] = ,
[m d(k)]2
k
where d(k) = .
1+k
(m)
Use the relationship given by (5.8) to find formulas for Yx .
As P (Kx = k) is a probability mass function, P (Kx = k) = k1 qx = 1. If we increase the
k=1 k=1
starting index by 1, that is, find k1 qx , we get
k=2
k1 qx = k1 qx 11 qx
k=2 k=1
= 1 qx
= px .
Therefore,
v 0 (qx + 1 qx + 2 qx + ) +v (1 qx + 2 qx + 3 qx + ) +v 2 (2 qx + 3 qx + ) + = v t t px .
t=0
px 2 px
k1 qx =1=0 px
k=1
This equation makes intuitive sense. If (x) survives to time t (with probability t px ), (x) receives a
payment of 1, which has present value v t . As a
x is the APV of a life-annuity due, (x) is guaranteed
one payment at the beginning of the year, so t = 0 is the starting index (with present value v 0 = 1).
Using the intuition we have developed, we can find that
ax = v t t p x , (5.28)
t=1
where t = 1 is the starting index, as the present value of the first payment received is v 1 .
5.6. kTH MOMENTS OF ANNUITIES 85
The intuition we have developed can also be extended to continuous annuities. We show that
x = v t t px dt.
a (5.29)
0
a
x = a
t fx (t) dt
0
= a
t t px x+t dt
0
d
= a
t ( [t px ]) dt.
dt
0
1 et et d
We approach using integration by parts. If u = a
t = , du = = et = v t . If dv = [t px ],
dt
v = t px . So
d
a
( [
t dt t xp ]) dt = a
t ( p
t x ) (t px )v t dt
t=0
0 0
0 0 px ] + v t t px dt.
at t px ) a
= [lim (
t
0
1
Intuitively speaking, lim t px = 0, as (x) cannot survive infinitely. Also, lim a t = .
t t
1
Hence lim ( at t px ) = (0) = 0. We also find that a
0 = 0 and 0 px = 1. So the term
t
at t px ) a
[lim ( 0 0 px ] = 0.
t
Hence we get (5.29). We could also find formulas for mthly annuities that follow a similar format,
but they are extremely messy and will not be presented in this text. The main point to get
from this section is that (x) will receive an annuity payment if (x) is alive at the time
the payment is made.
[1] Cunningham, Robin J., Herzog, Thomas N., and London, Richard L., Models for Quantifying
Risk, 4th edition (2011), ACTEX Publications.
[2] Cunningham, Robin J., Herzog, Thomas N., and London, Richard L., Supplement Note for
Candidates using Models for Quantifying Risk, Fourth Edition, 2012, ACTEX Publications.
[3] Dickson, David C.M., Hardy, Mary R., and Waters, Howard R., Actuarial Mathematics for Life
Contingent Risks, 1st edition (2009), Cambridge University Press.
[4] Dickson, David C.M., Hardy, Mary R., and Waters, Howard R., Supplementary Notes for
Actuarial Mathematics for Life Contingent Risks, 2011.
87