Findings on Social Security Systems and Mortality Projections

Findings on Social Security Systems and Mortality Projections

The Living to 100 Symposium series continues. This article highlights the connection between national population statistics and actuarial mortality projections for social insurance programs.

By Sam Gutterman

Expected mortality is simultaneously a significant factor in the projection of the long–term cost and future financial conditions of social insurance programs and a controversial one. This demographic assumption is particularly significant because public policy discussions are so reliant on the objective nature of these projections; for example, the American Academy of Actuaries has recently come out in favor of increasing the retirement age for Social Security reflecting expected changes in future mortality, a move that is being considered by most other countries around the world as the expected cost of their public retirement programs has been increasing. The experience provided and methodology used in developing the assumptions can provide helpful insight into mortality projections that are an integral part of many other fields of actuarial practice.

Two panels at the 2008 Living to 100 Symposium–sponsored by the Social Security Committee of the Society of Actuaries–addressed this topic. The bases for these projections as made in Canada, the United States and the United Kingdom was presented by leading actuaries intimately involved in the actuarial analysis of public social retirement programs. Three excellent papers were submitted to the symposium addressing this topic. They were written by Jean–Claude Ménard, FSA, FCIA, chief actuary of the Canadian Pension Plan (CPP); Alice Wade, ASA, EA, MAAA, deputy chief actuary of the U.S. Social Security Administration (SSA); and Adrian Gallop, FIA, actuary in the Government Actuary's Department of the United Kingdom. Danita Pattemore, FSA, FCIA (CPP); Al Winters, FSA, EA, MAAA (SSA); and Steve Goss, ASA, MAAA, chief actuary (SSA); also participated in these panels.

The development of mortality projections for social insurance programs involves rigorous studies of national population statistics. The identification and consideration of the current level of mortality and historical trends are important elements of these projection processes.

Of course, the derivation of long–range mortality assumptions is subject to quite a bit of difference of opinion. On the one hand, many biologists have claimed that significant future advances in extending the human life span may be on the horizon. In addition, those statisticians and actuaries who extrapolate from past trends–for example many who have applied the Lee–Carter method–have projected very significant future improvements. In contrast, some behaviorists have hypothesized that environmental and human behavior may result in adverse mortality trends, possibly offsetting other factors.

Population Experience

Since social insurance is provided to a large majority of a country's population, the starting point of all of these mortality projections is recent national demographic experience. All of the authors and presenters indicated reliance on their country's population data primarily developed from government agencies, with data quality problems experienced at the very old ages. These problems have been minimized in the United States by use of data from the Medicare system–the data from which has been deemed to be more accurate than public sources–as the population covered is very close to complete.

The two life expectancy graphs indicate the overall historical trend in life expectancy by gender at birth (Graph 1) and age 65 (Graph 2) for the United States, Canada and the United Kingdom. Both graphs include lines which represent the corresponding difference between female and male life expectancies. Several observations are clear: with few exceptions, the trends over the last century have seen a continual increase in life expectancy, with life expectancy at age 65 showing a significant increase over the last 30 or 40 years, depending on the country. These recent increases in life expectancy during the retirement ages–although showing somewhat inconsistent patterns after age 85 during some periods–have had a significant effect on the current and expected benefits payable under old age social insurance programs. In some respects, trends at these ages are the most difficult to project as well.

These graphs also show the changing nature of differences in life expectancy between the genders, with the overall trend being quite similar for the three countries. These differences increased over the first two–thirds of the last century, with decreases experienced over the last quarter or third of a century. A major contributor to this pattern has been changes in cigarette smoking prevalence for each gender over the last century, although it certainly isn't the only cause. As the effect of the decrease in smoking for females begins to emerge in population statistics over the next couple of years, the reduction in the difference between female and male life expectancy will likely decelerate. This example strongly suggests that blindly projecting mortality improvement by gender may not be appropriate. Known conditions and underlying trends should be considered.

Sources of Mortality Trends

Each of the three presented papers shows some of the significant drivers of both past national mortality experience and those that may drive future mortality. Those of the past century have included: significant reductions in infant and childhood deaths, improvement in sanitation, better public health measures, the introduction of vaccines and antibiotics, and improvements in the overall standard of living and education. In the last several decades, mortality has also benefited from reductions in cigarette smoking, improved access to medical care–due in part to the introduction of Medicare and Medicaid in the United States–and reduction in cardiovascular disease due in part to the mitigating effect of blood pressure and cholesterol medication in all three countries. But the relative effects of some of these historical drivers have decreased over the last few decades. The reduction in smoking prevalence for males that has occurred over the last few decades has certainly increased their life spans, with the effect on life span for females somewhat delayed as their change in smoking habits has occurred more recently.

Factors cited in the papers that may be significant in the future include:

  • Medical diagnostic, treatment and life–sustaining advances.
  • Translation of newly found genetic and aging information to medical drugs and treatment.
  • Incidence of violence, suicides and accidents.
  • Environment pollution, including air and water quality.
  • Amount of physical activity.
  • Nutritional trends.
  • Obesity and diabetes.
  • Smoking and drug prevalence.
  • Epidemics and natural disasters.
  • Education and wealth effects.
  • Changing demographic mix due to such factors as immigration.

Obviously, not all of these factors will have favorable effects on mortality. Some will have an immediate impact, while others can take a long time to influence mortality. It's important to note too that they can affect people of various ages and genders differently. For example, reductions in smoking prevalence can take more than 20 or 30 years to show up in overall mortality trends, while for females decreasing mortality trends from this source are only now beginning to be observed because their decreasing smoking prevalence started later than that of males. It remains to be seen how significant the effect of the exploding levels of obesity and diabetes in all three countries will be on health and life spans of the population in general. Trends are already emerging, indicating that obesity can result in adverse mortality consequences such as Type 2 diabetes, cardiovascular risk factors, various types of cancers, and kidney and gallbladders diseases.

Period Versus Cohort Effects

Certain mortality trends are period effects, while others are cohort effects. Examples of the former are epidemics and changes in many medical treatments, while examples of the latter are changes in nutrition of infants. The papers describing the Canadian and U.S. mortality projections indicate that, at least so far, mortality trends have been dominated by period effects. An example was the introduction of Medicare in 1965 in the United States. In contrast, cohort effects have also been quite noticeable in the United Kingdom. This could be due to changes in childhood nutritional patterns, introduction of the National Health Service and medical advances, with faster improvements experienced by the U.K. generation with births in the later 1920s and the 1930s than those born either earlier or later. Those born in the early 1950s and early 1960s are experiencing continuing lower rates of mortality improvement compared with those born in between these periods. These cohort differences have been reflected in United Kingdom social insurance mortality projections since the early 1990s.

But even though cohort effects have yet to be observed in North America, it is possible that different behavioral effects by generation may eventually result in cohort–related mortality in Canada and the United States.

Projection Methodologies

Mortality projections in the three countries are developed on the basis of rigorous study and regular review by experts. In the United States for example, mortality projections are subject to a quadrennial review by technical panels featuring well–known actuaries, economists and demographers and annually by the Trustees of the Social Security system and their staffs, while CPP projections are peer reviewed triennially by a panel of independent actuaries. In the United Kingdom, population projection assumptions are set after considering input from meetings with an advisory panel of demographic experts and from discussions with the principal users of the projections.

All three sets of the projections consider possible future sources of changes in mortality, separately by major age groupings and gender. In the United States, mortality rates from seven major categories of cause of death are separately projected over the short–term and explicitly projected based on recent trends in all seven categories of causes. Social insurance actuaries in Canada and the United Kingdom consider prospects by cause in each of the categories on an implicit basis for each of the major age groups, but don't model them separately. In the United Kingdom, as mentioned, separate projections are made by birth cohorts.

The underlying factor modeled is the age–gender–year rates of mortality improvement.

Relative mortality differentials by both benefit size and marital status are considered in the United States, as different population segments are expected to experience different levels of mortality.

Table 1 (on the previous page) shows the average historical and estimated annual mortality improvement by age grouping for the three countries, both in terms of long averages and the ultimate level assumed.


Before any projection is finalized, the social insurance actuary realizes that every mortality projection is only based on an average of expected scenarios and remains uncertain despite the best of efforts. As a result, it is highly important to provide easy–to–understand information as to the degree of this uncertainty to avoid the possible misconception by users that there is only one possible future path that mortality will take. The traditional way of providing this transparency is to provide alternative high and low projections–used by actuaries in all three countries–in addition to expected values. These alternative projections are provided for illustrative purposes and are not intended to provide a worst or best case scenario.

At the Living to 100 Symposium, Al Winters and Danita Pattemore elaborated on the use of stochastic mortality projections in the United States and Canada, respectively. Both used autoregressive integrated moving average (ARIMA) models, with historical experience as a basis of the uncertainty reflected. A fan–type projection results, an example of which is shown in Graph 3 for male life expectancy at age 65 from the United States. A key ongoing challenge associated with these stochastic projections is their effective communication so that policy decision–makers better recognize the uncertainty associated with these projections. In the United Kingdom, the Office for National Statistics (ONS) is currently developing a stochastic forecasting model and aims to publish a set of probabilistic projections–on an experimental basis–in 2009.

Review the Papers and Presentations!

This article is just an introduction to the topic of mortality projections and their impact on social security systems. Extensive details are provided in the papers and presentations, as well as on the respective Web sites of these social insurance programs. Visit the SOA Web site at to find this information.

Future mortality affects many actuarial practice areas and will be of even more significance in the future. The papers presented at the symposium illustrate approaches taken to project future mortality in the general population and in the social insurance practice area. Mortality is no longer a risk to be ignored.

Sam Gutterman, FSA, MAAA, FCAS, FCA, HONFIA, is director and consulting actuary at PricewaterhouseCoopers LLP in Chicago.