You can search for articles in back issues of Contingencies from July/August 2000 to March/April 2009 using the search box to the right. Simply type in subject words, author's name, or article title and click search. To search for articles from May/June 2009 to the present, go to the current digital issue of the magazine and use the search function on the left of the top navigation bar.
Turning the Tables:
Pension Tables Should Reflect Improving Mortality
by Emily Kessler
(Editor’s note: This is a longer, more detailed version of Emily Kessler’s article in the November/December 2005 print edition of Contingencies. Table II, Projected Mortality, is corrected from what was incorrectly printed in the online and print editions of the November/December 2005 issue of Contingencies.)
The 1983 Group Annuity Mortality table (GAM-83) is probably the most common mortality table used by pension actuaries; 75 percent of the plans in a 2003 Watson Wyatt survey of actuarial assumptions and funding used GAM-83 for funding calculations. However, there are three more recently published tables that warrant consideration for use in pension valuations: the 1994 Uninsured Pensioner Mortality Table (UP-94), the 1994 Group Annuity Mortality Table (GAM-94) and the RP-2000 Mortality Table (RP-2000).
Which is the most appropriate table to use? Should actuaries be moving to these new tables, or does GAM-83 still represent a reasonable expectation for most plans? What are the consequences if plans don’t use current mortality tables, and what options are open to actuaries?
The GAM-83 table was constructed after a review of GAM-71 and insurer experience showed that GAM-71 was inadequate. When GAM-83 was developed, there weren’t sufficient credible data available to construct a new table, so the developers used the same annuitant mortality experience on which GAM-71 was based. This is mortality experience from 1964 to1968.
Researchers then reviewed U.S. population statistics to determine mortality improvements from 1966 to 1975. They projected additional mortality improvements to 1983 based on 1966–1975 trends, and, according to the GAM-83 report, added a 10 percent margin for conservatism to add “a degree of safety to reserves. Since mortality experience varies by company, margins should insure that the mortality table can cover the lightest mortality experience of all companies except for truly exceptional cases.”
UP-94 & GAM-94
The UP-94 table is based on uninsured pensioner experience projected to 1994. It was developed by the Society of Actuaries (SOA) to replace UP-84 after a study of 1985–1989 mortality experience of 29 retirement systems found mortality rates were between 82 percent and 86 percent of those expected under UP-84.
Similarly, the GAM-94 table is based on group annuitant experience projected to 1994. This table was developed to replace GAM-83 after a study of 1986–1990 annuitant experience showed steady declines in ratio of actual to expected (GAM-83) mortality, particularly for males.
During the development of the two tables, recent experience for uninsured pensioners was compared to recent experience for group annuitants, and no significant mortality difference was detected. Researchers compared mortality rates at ages 66–95 for group annuitants and for participants in the federal Civil Service Retirement System (CSRS), uninsured plans (24 private and one state), and the Railroad Retirement System.
Researchers found that the group annuitant and uninsured pensioner rates were quite similar (the Railroad Retirement System showed higher mortality rates). Thus, the same underlying data were used for both the GAM-94 and the UP-94 tables. Both tables were constructed as follows:
The only difference between the final UP-94 and the GAM-94 tables is that the GAM-94 table includes a 7 percent margin. GAM-94 was designed for insurance reserves, which need margins for deviations in blocks of business. For an insurance company, a 5 percent margin provides a 95 percent confidence level on 3,000-life block of business. The additional 2 percent margin was added to account for variations in types of work (white-collar or blue-collar), income, and geography. Also, according to the GAM-94 report, it was felt that an additional margin would be needed as tables are adopted by state insurance commissioners and aren’t changed often thereafter.
The RP-2000 table is the only table whose underlying rates are based solely on retirement plan mortality experience. It was developed by the SOA specifically for current liability calculations. The Retirement Protection Act of 1994 (RPA 94) allowed the Secretary of Treasury to promulgate a new table for current liability purposes in 2000. The SOA conducted a study of uninsured pension plan mortality to ensure that the Treasury Department would have current information available when considering updating the table. As there was no current table based on uninsured pension data (UP-94 was based partly on group annuity experience) a decision was made to conduct a separate study.
In the construction of the RP-2000 table, data were collected from private employers (those affected by current liability provisions) for plan years ending 1990 through 1994. From the data collected, rates were constructed for employees 30 to 70 years old, healthy annuitants 50 to 100, (based on a blend of retiree and beneficiary data), and disabled retirees 45 to 100. For ages below 30, rates were extended to blend with UP-94, and for ages above 100 they were projected to a value of 0.40 at age 106 for males and 115 for females, with an ultimate rate of 1.0 at age 120.
Rates were adjusted for mortality improvement from 1992 to 2000 using the data underlying Actuarial Study No. 110 and federal CSRS data. The AA scale that had been published with the UP-94 and GAM-94 reports was also published with the RP-2000 report. The committee felt that this scale was reasonably close to what was seen in Social Security trends and consistent with other groups’ data. Although minor adjustments could have been made, they were not considered significant enough to justify a new scale.
To illustrate how the tables compare, Table 1 shows the change in annuity values, deferred to age 65 (immediate if over age 65) for various ages.
* Deferred annuity (to age 65) for ages below 65; immediate annuity for ages 65 and above
Female mortality rates haven’t decreased; they’re slightly higher in the newer tables than in GAM-83. Partly, this is due to the 10 percent margin in GAM-83 (female mortality didn’t improve as much as was expected). This also reflects that GAM-83 female mortality is based on relatively little actual experience.
What does this mean for the practicing actuary? Is the GAM-83 mortality table still a reasonable mortality table for use in valuation? When considering these questions, we must look to actuarial standards of practice for guidance.
Actuarial Standard of Practice 35, Selection of Demographic and Other Noneconomic Assumptions for Measuring Pension Obligations, gives very clear, specific guidance on the selection of mortality tables, as well as other demographic assumptions. (Go to www.actuarialstandards.org to read the ASOP; the following limited excerpt in no way is intended to be a substitute for the ASOP’s complete text.) Here’s what ASOP 35 says:
— “the likelihood and extent of mortality improvement in the future”
— “the use of a different mortality assumption for disabled lives”
— “the use of different mortality table for different participant subgroups and beneficiaries.”
We’ll address mortality projection later. Let’s go back to the question of whether GAM-83 is an appropriate mortality table, based on the guidance provided by ASOP 35.
First, in no way is anyone saying that GAM-83 is never an appropriate table. There are certainly circumstances in which GAM-83 will be the best choice based on the appropriate assumption universe. And there are probably circumstances in which other older, less conservative tables (GA-71, UP-84) are still appropriate.
Can it be argued that GAM-83 is the appropriate table for most plans? Consider what we know, 20 years after the publication of GAM-83:
Absolute mortality rates have changed and they’ve improved (or not improved) differently for males and females; and, for each gender, differently by age. In other words, GAM-83 probably doesn’t represent, for most populations, the correct level of mortality, and even if projected, probably won’t reflect the right pattern of mortality. We shouldn’t be surprised; GAM-83 is based on mortality experience from the 1960s projected to 1983, and on mortality improvement trends from the 1960s and 1970s, with the addition of a 10 percent margin.
We recognize that GAM-83 is prescribed for the current liability calculation. And ASOP 35 notes that, when an assumption is prescribed, the actuary is obligated to use it for the purpose for which it was prescribed (paragraphs 2.6, 3.8).
But just because a particular assumption is prescribed in one calculation, ASOP 35 doesn’t say that it therefore becomes the most appropriate assumption for all the other calculations. And for good reason. Consider the following example, using the prescribed mortality for current liability and a completely hypothetical situation:
This is an exaggerated example, but drawn to make a point: We use prescribed assumptions when they’re prescribed, sometimes criticizing under our breath those who prescribed them. But just because they’ve been prescribed in one circumstance doesn’t make them the best assumption in another. It doesn’t mean they aren’t, but it doesn’t mean they are.
To Project or Not to Project?
It’s not a question; it’s part of your assumptions. Mortality table construction has changed over the past 20 years. When the GAM-83 tables were created, computing systems were limited. Tables were built with substantial margins to allow not only for variation in experience, but also because programming in a new table took significant effort. And static or generational projections were rare.
However, the construction of recent tables has reflected updates in our systems and our abilities to create individual projections. The three most recent tables reflect only mortality improvements through their creation dates (1994 for UP-94 and GAM-94, 2000 for RP-2000). This is because their creators expected users to make explicit assumptions about mortality improvement. So every time actuaries use one of these tables, they must make an explicit decision about whether and how to project mortality improvements beyond the table date. In other words, by not making projections, the actuary has made the explicit decision not to assume any future mortality improvements beyond the date of the table’s creation.
In a report on choosing between UP-94 & GAR (group annuity reserving)-94, which coincided with the publication of the UP-94 and GAM-94 tables, the actuaries responsible for their creation noted:
Though the incorporation of explicit projection scales has not previously been standard practice in the valuation of retirement plans, we believe that actuaries should carefully consider using mortality trend projection if adopting a version of the UP-94 Table. The argument for use of mortality improvement trends is bolstered by the following observations:
Similarly, in its issuance of the RP-2000 report, the committee that developed the table recommended that
... in the view of the long history of improvement in non-disabled mortality rates in all of these sets of data, pension valuations should take trends in long-term mortality improvement into account. From a theoretical standpoint, the [committee] believes that the use of generational mortality improvement, as in the GAR-94 table, is an appropriate way of reflecting this improvement. In cases where it is not material or cost effective to incorporate generational mortality improvement into a calculation, the actuary should project mortality improvement on a comparable static basis.
And finally, a recent SOA-commissioned study shows the effects of not taking mortality improvement into account. The paper examined, theoretically, what would happen to a sample plan’s funded status, contributions, and FAS 87 expense (among other measures) given known mortality improvements and different actuarial assumptions, which tracked or lagged actual mortality improvements to varying degrees.
The study by David F. Kays, which defined “ideal” assets as sufficient to cover actual mortality improvement, found that “periodically updating the mortality table assumption to reflect current mortality levels with or without mortality scale projections or using mortality projected beyond the valuation date may accumulate assets closer to the ideal assets. Consistently using mortality tables that are not current will eventually accumulate assets less than ideal.” In particular, when considering whether assets were sufficient to cover ABO with actual mortality improvement, the study said, “the greater the mortality assumption is from the experience, the less chance assets will be large enough to cover [ABO].”
In other words, for assets to accumulate to a relatively level percentage of their ideal, the mortality assumption ought to be updated periodically, and at least projected to the valuation date by the appropriate mortality improvement scales. Tables that were projected beyond the valuation date did a better job of approximating a generational table; the ideal projection point would likely vary by plan population. But as we’ve already seen, in some cases the new factors showed higher mortality than existing tables. Is there really a need to project specific improvements onto the tables?
Note that none of these most recent tables—UP-94, GAM-94 and RP-2000— would be considered to reflect current mortality experience, unless there has been no improvement in mortality between their creation dates and today. We have some evidence that mortality has improved over the past 10 years. (The creators of the RP-2000 table, which was created with data gathered six years after the UP-94 data, noted that there appeared to have been an improvement in mortality in that period that tracked Projection Scale AA; i.e., their year 2000 table tracked well against the UP-94 table with six years’ projection. These tables were based on two different populations, however, so this is a hypothesis, rather than a conclusion.)
If the tables are simply brought up to date — from their creation dates in 1994 and 2000, respectively, to 2005 — the ratio of the differences in annuity factors between GAM-83 and the projected tables comes much closer together for females, and widens even more for males. And if full generational improvements are reflected, then the mortality differences are much wider. Table 3II shows selected rates with improvement.
* Deferred annuity (to age 65) for ages below 65; immediate annuity for ages 65 and above
Note: for the generational annuity factor, the sample annuitant is assumed to be age X in 2005. The factor of 2.80 for a 35-year-old male is for a male who is 35 in 2005. The factor of 13.14 for a 65-year-old female is for a female who is 65 in 2005. Table II, Projected Mortality, is corrected from what was incorrectly printed in the online and print editions of the November/December 2005 issue of Contingencies.
So what should actuaries consider when projecting mortality? The committee that oversaw the UP-94 and GAM-94 tables recognized that many factors influence decisions to project mortality:
Does this mean you’re always required to project mortality, and more critically, are you required to use generational mortality? Not necessarily. Each population is different. That’s where your actuarial professional judgment comes in.
Food For Thought
Let’s assume a pension plan has a normal retirement age of 65, unreduced early retirement at 62, and early retirement reduction factors of 5 percent per year before age 62. (The benefit paid at age 55 is 65 percent of the normal retirement benefit). The actuary currently uses 1983 GAM mortality, retirement rates of 5 percent per year before age 62, 50 percent at age 62, 5 percent at 63 and 64, and 100 percent at age 65. The weighted average retirement age is 61.8. Consider three changes to mortality: to RP-2000, to RP-2000 projected to 2025 and to RP-2000 generational, for each of three sample male participants, a 65-year-old (born in 1940), a 45-year-old (born in 1960), and a 25-year-old (born in 1980).
Do these assumptions seem reasonable? Consider what the change in mortality does to life expectancy:
Our sample plan, like many other plans, was designed to help move the war generation out of the work force to make way for the baby boomers. It has provided subsidized early-retirement benefits for anyone wishing to leave the workforce before age 62—subsidies worth as much as 30 percent at age 55. Our actuarial assumptions reflect that prior generations have, and future generations probably will, continue to take this early retirement subsidy.
A man who expects to live to 82 may reasonably be expected to retire at 62, particularly when there are generations of workers ready to take his job. But is it reasonable to expect that someone born in 1980, who with improved mortality would have a life expectancy of 87, to also retire at age 62? If improvements in life expectancy also bring improvements in health at older ages, might our disability rates at older ages (e.g., 50-plus) also decrease?
When projecting mortality, all things must be considered in balance: If mortality improves, what will happen to disability rates? Will retirement ages increase as people work longer, either out of necessity or desire? It’s not all that simple. You need to use your actuarial professional judgment.
Emily Kessler is the staff fellow, retirement systems, at the SOA and acts as SOA liaison to the Academy’s Pension Practice Council and Pension Committee. The views of the author are her own, and are not those of the American Academy of Actuaries, Academy, Actuarial Board of Counseling and Discipline, Actuarial Standards Board, SOA, or the actuarial profession.
American Academy of Actuaries, letter to the Department of Treasury on RP-2000, April 2001.
American Academy of Actuaries, letter to the IRS on RP-2000, December 2003.
Kays, David F., Impact of Mortality Table Projection Scales on Defined Benefit Pension Plan Valuations, Society of Actuaries, 2005.
Society of Actuaries (SOA) Group Annuity Valuation Table Task Force, 1994 Group Annuity Mortality Table and 1994 Group Annuity Reserving Table, Transactions of the SOA 1995, volume 47,865 – 919.
SOA, 1999 Annual Meeting discussion of RP-2000 table and projecting mortality.
SOA Committee on Annuities, Development of the 1983 Group Annuity Mortality Table, Transactions of the SOA 1983, volume 35, 859– 899.
SOA Committee on Retirement Systems Research and Committee on Retirement Systems Practice Education, The UP-94 and GAR-94 Tables: Issues in Choosing the Appropriate Table, Transactions of the SOA 1995, volume 47, 795 – 817.
SOA Retirement Plans Experience Committee, RP-2000 Mortality Tables, 2000.
SOA UP-94 Task Force, 1994 Uninsured Pensioner Mortality Table, Transactions of the SOA 1995, volume 47, 819 – 863.
Watson Wyatt, 2003 Survey of Actuarial Assumptions and Funding, 2004.
Contingencies (ISSN 1048-9851) is published by the American Academy of Actuaries, 1100 17th St. NW, 7th floor, Washington, DC 20036. The basic annual subscription rate is included in Academy dues. The nonmember rate is $24. Periodicals postage paid at Washington, DC, and at additional mailing offices. BPA circulation audited.
This article may not be reproduced in whole or in part without written permission of the publisher. Opinions expressed in signed articles are those of the author and do not necessarily reflect official policy of the American Academy of Actuaries.