3. WITH IMPROVED GLOBAL MORTALITY CONFIRMED, THE QUESTION BECOMES, “WHAT
ADJUSTMENTS DO ACTUARIES NEED TO MAKE?” THIS ARTICLE ANSWERS THAT AND
OTHER QUESTIONS. BY MARIANNE PURUSHOTHAM
O Population Life Expectancy at Birth by Country
VER THE LAST CENTURY, gen-
eral population mortality has im-
proved significantly and on a global
MALES
basis. As this trend appears likely to continue
at least over the near term, mortality improve- India
ments are expected to continue to impact the
viability of national social programs worldwide Japan
as well as the financial stability of insurance pro-
Australia
grams, including pension plans and annuities.
2008
Switzerland 1990
As a result of the current trends in mortality
Spain 1950
improvement, insurers have begun to incor-
1900
porate longer term assumptions regarding
England/Wales
mortality improvement in both pricing and
financial projections, resulting in greater Canada
scrutiny from both an audit and regulatory
perspective. United States
To assist actuaries in developing mortality im- 0 10 20 30 40 50 60 70 80 90
provement assumptions for various product
lines including individual annuities, pension, FEMALES
life insurance and long-term care insurance,
the Society of Actuaries (SOA) commis- India
sioned a research project to compile both his-
Japan
torical data regarding mortality improvement
experience as well as information regarding Australia
techniques for modeling future assumptions. 2008
This article provides a general overview of in- Switzerland 1990
formation included in that report. 1950
Spain
1900
MORTALITY IMPROVEMENT
England/Wales
EXPERIENCE
Population Data Canada
From 1900 to 1950, enormous strides were
made in improving mortality rates on a global United States
basis, particularly at ages under 35. This was
largely due to an expanded focus on public 0 10 20 30 40 50 60 70 80 90
Source: U.S. Census Bureau Estimates from Various Databases, American Journal of
Clinical Nutrition, Changes in Life Expectancy 1900-1990, published 1992 and World
Health Organization, Statistics 2010.
Males Females
AUGUST/SEPTEMBER 2011 | THE ACTUARY | 21
4.0%
4. health initiatives in many countries that led to tality between specific ages. Here we discuss experienced a decline in improvement
the introduction of large-scale immunization improvements in mortality during key life levels during the current decade. The
programs, penicillin and sulfate drugs, and stages for four of the more developed coun- most notable exception to this trend is
other disease eradication methods. tries (United States, Canada, United Kingdom Australia, where young adults exhibited
and Australia) during the period 1940 to 2007. a notable increase in mortality improve-
Since 1950, both in the United States and ment (nearly 4.9 percent per year for
across the globe, mortality improvement and The following trends are noted: males and 3.1 percent per year for fe-
resulting increases in life expectancy have males) during the period 2000 to 2007.
continued, but at a slower pace. The most re- • nfants (the rate of mortality between
I
cent data from the World Health Organization ages 0 and 1): Rates of mortality im- • iddle Ages (the rate of mortality
M
indicates that male and female life expectan- provement for infants have slowed in between the ages of 35 and 65): Indi-
viduals aged 35 to 65 have experienced
UNDERSTANDING THE SOURCES OF PAST IMPROVE- moderate but steady levels of improve-
ment in mortality since 1960 (generally
MENTS IN MORTALITY CAN BE VALUABLE IN THE between 0.5 and 2.5 percent per year).
PROCESS OF DEVELOPING ASSUMPTIONS REGARD-
• etirement Ages (the rate of mortal-
ING THE FUTURE. R
ity between the ages of 65 and 85): In-
dividuals between ages 65 and 85 have
cies at birth are nearing 80 and 85 respectively recent years in all four countries, from exhibited lower levels of improvement,
for many developed nations. highs averaging between 3- and 5-per- with the largest increases occurring be-
cent per year between 1940 and 1980 to tween 2000 and 2007 (between 1- and
FACTORS AFFECTING POPULATION rates generally less than 2 percent dur- 2-percent per year).
MORTALITY IMPROVEMENT LEVELS ing 2000to 2007.
Understanding the sources of past improve- • ld Age (the rate of mortality between
O
ments in mortality can be valuable in the pro- • hildren (the rate of mortality between
C the ages of 85 and 100): Since 1940, the
cess of developing assumptions regarding the ages 1 and 20): Greater mortality im- population age 85 and older has exhibited
future. Therefore, the SOA report also examined provements were exhibited by this group little or no improvement and even some
the correlation between mortality improvement during the period 1940 to 1960 (between deterioration in mortality experience.
levels and various demographic factors. 3- and 8-percent per year) as children This observation has led some to theorize
and young adults benefited from the that in the future, mortality improvements
Age impact of stronger public health initia- are likely to be focused at the older ages
Between 1940 and 2006, improvements in tives. Improvements also appear to have where fewer strides have been made to
mortality were strong for most age groups, picked up between 1980 and 2000, pos- date. And, there is evidence for increas-
including infants, children, sibly due to a decrease in the number of ing improvement levels over the current
young adults, the middle deaths from accident or injury. decade for the oldest ages.
ages, retirement and old
age. • Young Adults (the rate of mortality Gender
between ages 20 and 35): Rates of mor- In the United States and Canada between
One tool in the analysis of tality improvement for young adults are 1980 and 2000, male improvements out-
changes in life expectancy also driven to a large extent by lower paced female improvements by an aver-
at birth is the examination rates of accident and injury. With a few age of 0.5 percent per year. In the United
of changes in rates of mor- exceptions, this age demographic has Kingdom, males experienced greater
22 | THE ACTUARY | AUGUST/SEPTEMBER 2011
5. Switzerland 1990
Spain 1950
1900
England/Wales
MORTALITY
IMPROVEMENTS
Canada
United States
status are highly correlated with mortality
Differences in Mortality Improvement by Level
0 10 20 30 40 50 60 70 80 90
and mortality improvement experience. The
wealthier, more highly educated and mar-
of Education Achieved ried populations tend to exhibit lower levels of
U.S. WHITE POPULATION mortality and also appear to have experienced
larger levels of mortality improvement. As an
Males Females
example, for the U.S. white male population
4.0% ages 25 to 64, the results of a 2008 study of ex-
perience between 1993 and 2001 shows that
3.0%
mortality improvements for those with 16 or
more years of education were approximately
2.0%
twice those of the population as a whole.
1.0%
LIFE INSURANCE DATA
0.0%
For life insurance in particular, actuar-
-1.0% ies have less consistent data on which to
base their views regarding future mortal-
-2.0% ity improvement levels, especially over the
longer term. Care needs to be taken in in-
-3.0%
terpreting insured data as changes in select
-4.0% mortality over time may be impacted more
Less than 12 yrs 12 yrs 13-15 yrs 16+ years All Levels by changes in the industry including shifts
in the target market, distribution methods,
Source: Widening of Socioeconomic Inequalities in U.S. Death Rates, 1993 to 2001,
Ahmedin Jemal, Elizabeth Ward, Robert N. Anderson, Taylor Murray, Michael J. underwriting processes or risk classification
Thun, published 2008. structures than true improvements in life
insured mortality. Data regarding improve-
ments in ultimate mortality may be more
levels of improvement for ages over 45 Geographic Region reliable since a major factor, the impact of
while females experienced greater im- There is data to support some variation in underwriting changes over time, should be
provements for ages under 45. And dur- mortality improvement experience by geo- greatly dampened.
ing the most recent experience period graphic region for the United States, Canada
examined, 2000 to 2007, the pattern of and the United Kingdom, with the lowest Utilizing the results of regular periodic studies
gender differences has become blurred levels of improvement generally seen in the of individual life insurance mortality experi-
for all three countries. As a result, both more rural and lower-income areas of the ence produced by national actuarial organi-
the September 2010 Canadian Institute of countries. These differences were generally zations over the past several decades, chang-
Actuaries report1 recommending mortal- small for the experience periods and coun- es in mortality rates for life insureds can be
ity improvement levels for individual life tries examined. calculated for the select and ultimate period
insurance and annuity products and the for the United States, Canada and the United
most recent U.S. Social Security Admin- Socioeconomic Status Kingdom.
istration Technical Panel report 2 assume The results of several population-based stud-
that differences in mortality improvement ies support the view that socioeconomic Focusing on ultimate period mortality, we note
levels by gender will be eliminated over factors such as wealth, income level, high- the following trends in improvements versus
the next few decades. est attained level of education, and marital the general population results.
AUGUST/SEPTEMBER 2011 | THE ACTUARY | 23
6. United States population mortality along with some element ercise. There have been multiple and
For the experience period between 1978 of professional judgment. However, in today’s extensive applications of predictive
and 2006, the pattern of mortality improve- environment where mortality improvement modeling in the property-casualty insur-
ment by age and gender is quite similar assumptions can have a material impact on ance industry for experience analysis,
between the insured population in the ul- financial results, actuaries are beginning to fo- pricing, underwriting, claims manage-
timate period and the general population. cus greater attention on the analysis and review ment and strategic planning purposes.
Insureds and the general population exhib- of past experience, the factors that have influ- Through application of predictive mod-
ited similar age patterns of mortality, with enced that experience, and more robust meth- eling, the impact of each of these factors
insured mortality improvements slightly ods for developing future assumptions. on mortality or mortality improvement
greater than the general population on an is identified, and a model is developed
amount basis and slightly lower than the The academic community has made impor- that estimates future values as well as
general population on a number of policies tant strides in advancing modeling and projec- the probability of a specified outcome
basis. tion techniques for mortality and mortality im- (for example, future mortality improve-
provements over the past several decades and ment levels within a specified range)
Canada a brief overview of these techniques follows. under various conditions.
In Canada between 1972 and 2007, life
insureds in the ultimate period also expe- First, mortality projection models generally • xtrapolative projection techniques.
E
rienced levels of mortality improvement fall into one of the following basic categories. This category of projection models in-
similar to or greater than the population in cludes the earliest methods used by ac-
general at the typical insurance-buying ages • redictive modeling. A predictive
P tuaries, economists and demographers
(35 to 84) on an amount basis. model begins with the identification in setting future mortality improvement
of a group of factors (predictors) that assumptions based on the projection of
United Kingdom can influence future results for a given past trends that were identified in the
In the United Kingdom between 1980 and value being estimated, e.g., mortality, historical data.
2000, greater improvement levels were seen mortality improvement or other items
for permanent assureds (whole life and en- of interest. Focusing on life insurance
A typical extrapolative approach in-
dowments policies) than the population in mortality improvement as an exam- volves fitting a model to actual mor-
general for males, while female permanent ple, many of the predictors available tality experience for each individual
assureds exhibited rates of mortality improve- for traditional experience analysis calendar year in the experience period
ment more similar to the population in gener- (e.g., age, gender, product type, smok- available. The parameters underlying
al. Note that in the United Kingdom, mortality er status, risk classification) can be the models for each calendar year are
data is only available on a number of policies included in a predictive modeling ex- then plotted to identify significant pat-
basis at the current time.
MORTALITY IMPROVEMENT WANT MORE INFORMATION?
PROJECTION TECHNIQUES
Today, insured mortality im- This article is based on the SOA sponsored research paper, ”Global
provement assumptions Mortality Improvement Experience and Projection Techniques” by Mar-
are often developed ianne Purushotham, Emil Valdez, and Huijing Wu. It can be found on the
from an extrapo- SOA website at::
lation of past
trends in
24 | THE ACTUARY | AUGUST/SEPTEMBER 2011
7. MORTALITY
IMPROVEMENTS
terns that can aid in the determination ment of disease that may have led to
of future parameter values for projec- the death, including the impact of co-
tion of future experience. morbidities. In this modeling context,
better estimates of future mortality can
G
ompertz and Weibull are well-known be produced by considering the pro-
extrapolative models for mortality pro- gression of disease as well as potential
jection over time due to their relative interactions between different existing
simplicity. The Gompertz model, for morbidities. However, disease-based
example, projects mortality as the sum models are quite complex, typically
of an attained age dependent compo- involving multi-state transitions and
nent and an attained age independent
component. The Lee-Carter model and
A LIMITATION OF BOTH ALL-CAUSE AND CAUSE-
its many variations have also been used
extensively on a global basis, and Lee- SPECIFIC MORTALITY MODELS IS THAT ANY DATA
Carter is one of the methods employed IS RESTRICTED TO THE DEATH ITSELF.
by the Social Security Administration
in the United States for modeling future • ause-of-death specific models. To
C corresponding probabilities of transi-
levels of mortality improvement. Lee- date, the more common practice has tion that require detailed, longitudinal
Carter is a stochastic mortality projec- been to apply one of the models dis- data regarding the treatment and pro-
tion model based on the identification cussed earlier to all-cause insured or gression of various diseases. This type
of both the impact of attained age and population mortality experience to de- of information has been obtained in
calendar year. velop projections of future results. How- the United Kingdom through access
ever, it has also been argued that future to patient health databases supporting
• elational or targeting methods.
R public health care programs. Howev-
all-cause mortality experience can only
The basic approach for relational pro- er, in the United States, privacy laws
be reasonably understood if trends in
jection methods is to develop a func- make this type of data difficult, if not
specific causes of death and future likely
tion F that relates mortality in a specific impossible, to obtain and therefore
changes are understood. In theory, any
population under study to mortality disease-based models have not been
of the models discussed here can be
for a reference population. The devel- used extensively to date.
applied to the mortality experience for
opment of the function F is typically
specified cause-of-death groupings and
dependent on testing the fit of the rela- Bearing in mind the current limitations of in-
then combined in some manner across
tionship between the population being sured data as compared to population data as
all the groupings to produce aggregate
modeled and the reference population. well as the advantages and disadvantages of
assumptions. For these models, limiting
the various modeling approaches in current
the number of major groupings by cause
eference populations could include
R use, an industry level effort of the following
of death has been important in maintain-
long-term historical averages for a speci- form could produce valuable information to
ing a sufficient level of credibility.
fied country (e.g., low mortality coun- aid actuaries in their efforts to monitor trends
tries such as Sweden and Japan are and set assumptions regarding the future.
• isease-based models. A limitation
D
often used in this capacity). Another ap-
of both all-cause and cause-specific
proach is to assume that future mortality 1. onstruct and continually refresh a de-
C
mortality models is that any data is re-
improvements trend from their current tailed database of information regarding
stricted to the death itself. Therefore,
levels to a long-term target assumption the past and current impact on mortality
there is no consideration of the poten-
such as 2- to 2.5-percent per decade. of various key factors including lifestyle
tial impact of the timing and develop-
AUGUST/SEPTEMBER 2011 | THE ACTUARY | 25
8. MORTALITY
IMPROVEMENTS
•
Changes in levels of disease incidence global economy and the insurance industry
including emerging new diseases— and will require greater attention by actuar-
Collect data regarding new diseases ies. This approach allows for the application
and disease types as well as growth or of a rigorous modeling methodology in the
decline in current disease diagnoses. development of a baseline improvement as-
sumption, while retaining the element of pro-
•
Access to medical care—Follow the fessional judgment that forms the basis of cur-
development of the health care re- rent approaches and allows for incorporation
form initiative in order to gauge its of reasoned views regarding the uncertain
potential impact on future mortality future. A
and mortality improvement levels.
Marianne C. Purushotham, FSA, MAAA, is a senior
2. elect a sound modeling and projec-
S consultant at Towers Watson. She can be contacted at
trends and behaviors, past and recent marianne.purushotham@towerswatson.com.
tion methodology, preferably one that
medical advances, and demographic/ provides a reasonable fit to recent his-
ENDNOTES:
societal changes utilizing both insur- torical experience. In determining the 1
“Mortality Improvement Research Paper,” Canadian
ance industry data and other outside experience basis for model parameter
Institute of Actuaries, September 2010. http://www.
sources. This information will allow for development, due consideration will actuaries.ca/members/publications/2010/210065e.pdf
a better understanding of the potential need to be given to balancing the need 2
“Technical Panel Report on Assumptions and
impact of changes in these areas on cur- for a sufficient number of years to reduce Methods (2007),” Report to the Social Security
rent and future projections. “noise” in the average annual rates with Advisory Board, Washington D.C., October 2007.
the need for the model to reflect what http://www.ssab.gov/documents/2007_TPAM_
A partial list of factors that are expected REPORT_FINAL_copy.PDF
has occurred in the recent past.
to impact future mortality improvement
levels includes the following: 3. odify the initial results from application
M
of the projection technique developed
•
Medical advances—Collect data re- in 2 (above) by applying actuarial and
garding new advances in health care other expert judgment regarding future
that are expected to impact mortality expectations as to changes in the factors
improvements (newly introduced di- discussed in 1 (page 25) and their impact.
agnostic tests, treatments, phamaceu-
ticals). Consider the potential reduc- Some areas that are being closely watched
tion in mortality and the population by academicians and the medical commu-
that will be impacted and the period nity include obesity levels, smoking habits,
of time over which the reduction is and genetic indicators. And there is some
expected to emerge. evidence to support potential slowdowns in
mortality improvement in the United States
•
Lifestyle changes—Track levels of over the next several decades as a result of
smoking (number of new smokers health impacts of increased obesity levels.
and rates of quitting in adulthood),
alcohol intake, and levels of obesity Mortality improvement now has a material im-
and expected impact on mortality. pact on financial results in many areas of the
26 | THE ACTUARY | AUGUST/SEPTEMBER 2011