Summary
Wilmoth, writing as a member of the 2003 Technical Panel on Assumptions and Methods, audits the Social Security Administration's (SSA) Office of the Chief Actuary (OACT) mortality projection methodology for the 2003 Trustees Report (TR2003). He identifies five methodological choice points where OACT's procedure is either empirically unjustifiable or internally inconsistent. The central finding — dubbed here "every choice pessimistic" — is that TR2003's intermediate (central) scenario requires selecting the longevity-pessimistic option at every one of the five decision points, a joint probability that strains credibility as a central forecast. Using Human Mortality Database (HMD) cross-national data, Wilmoth shows that the US old-age mortality stagnation driving OACT's pessimism is an outlier among high-income countries and more likely reflects a transient deviation than a permanent structural break.
Key Claims
- 70-parameter structure: OACT selects 70 "ultimate rates of decline" (5 age groups × 2 sexes × 7 cause categories). These are derived primarily from 1979–2000 data (21 years) and approved by political-appointee trustees without publicly documented rationales. Example: OACT assumes cancer mortality declines 0.5%/year, but observed 1979–2000 rates were −0.06%/year for men and −1.13%/year for women — one assumption is off by a factor of 8 and in the wrong direction relative to the other.
- Cause-of-death forecasting creates hidden deceleration: When all-cause mortality is derived by aggregating cause-specific forecasts, the slowest-declining cause mechanically gains share over time, automatically slowing the projected total decline rate. The 2003 and 2007 Technical Panels both recommended eliminating cause-specific projection from the central forecast and retaining it only for analytical decomposition.
- Slope vs. endpoint method: Two methods for computing historical trend rates of mortality decline diverge importantly above age 80. The slope method (ordinary least squares (OLS) regression of log death rate on year, full period) is approximately equivalent to the Lee-Carter model. The endpoint method (ratio of final-period to initial-period rates) is approximately equivalent to OACT's piecewise approach. For ages 65–79, both methods agree closely for 1950–2000. For ages 80+, the endpoint method yields ~0.84%/year decline vs. the slope's ~0.98%/year — because the 1990s witnessed a US-specific stagnation in old-age mortality improvement that the endpoint method overweights and the slope method smooths across the full 50 years.
- "Every choice pessimistic" (Table 7): At five sequential methodological decision points — (1) historical baseline period, (2) slope vs. endpoint method, (3) cause-specific vs. all-cause projection, (4) ultimate rate specification, (5) convergence trajectory — TR2003's intermediate scenario consistently selects the option associated with lower projected longevity. Each individual choice might be defensible, but the consistent alignment toward pessimism across all five is not credible as a central forecast.
- US as international outlier: Using HMD data for 15 high-income countries over 1980–2000, 11 of 15 experienced accelerating old-age mortality decline. The US was one of only four deviants showing stagnation. This international comparison supports projecting a US recovery toward the international trend rather than anchoring to the anomalous 1990s period.
- Quantitative benchmarks: TR2003 projects e0 (life expectancy at birth) =82.8 years by 2070 (intermediate); the 2003 Technical Panel's alternative assumptions imply e0=84.4 by 2070; the 1999 Technical Panel's assumptions imply e0=85.2. The OACT's assumed ultimate rate of decline for ages 65+ is 0.68%/year — below the observed 20th-century average of 0.78%/year. TR2003 Long-Term Actuarial Balance (LTAB) = −1.92 percentage points (pp).
- Panel LTAB impact decomposes to near-zero net: The 2003 Technical Panel's recommendation on migration assumptions reduces the LTAB by 0.25 pp (more migrants → more workers → better solvency). The Panel's recommended mortality changes increase the LTAB by 0.24 pp (more longevity → more benefit years). The two effects approximately cancel: net Panel LTAB impact ≈ 0.
Concepts Introduced or Extended
Entities Mentioned
Quotes
"The official [TR2003] intermediate projection of U.S. mortality consistently relies on pessimistic assumptions about future rates of improvement — 'pessimistic' in the sense that each methodological choice tends to produce lower projected rates of improvement than the most defensible alternative."
"Eleven of 15 high-income countries experienced accelerating rates of decline in old-age mortality during 1980–2000. The United States was one of four exceptions."
My Take
This paper is the most methodologically precise audit of OACT's process in the literature. Its key contribution is the "every choice pessimistic" demonstration — not any single methodological decision is indefensible in isolation, but the systematic alignment across five choices toward the pessimistic option is a red flag about the process. The international comparison is particularly persuasive: the US 1990s mortality stagnation looks like a temporary deviation from a robust international trend, not a structural break that should anchor long-run forecasts. The near-zero net Panel LTAB impact (−0.25 pp migration + 0.24 pp mortality) is an important quantitative fact for policy discussions — it means adopting more defensible methodology roughly leaves the solvency measure unchanged, which defuses political resistance to methodological reform. The weakness of this paper is that it diagnoses the problem without proposing a fully specified alternative model — for that, Lee-Carter (or Soneji-King for behavioral covariates) is needed.