Summary
A comprehensive synthesis of the Lee-Tuljapurkar (LT) stochastic population forecasting program, covering methodological critique of official forecasts, a new attack on both standard and "random scenario" uncertainty methods, and a first full set of stochastic fiscal projections for the U.S. federal budget. The paper extends the 1994 JASA paper to fiscal applications and provides the most detailed published account of why Census and SSA uncertainty intervals are internally inconsistent across demographic quantities. Circulated as a working paper in 1998–1999 and published as a chapter in Alan J. Auerbach and Ronald D. Lee (eds.), Demography and Fiscal Policy (2000), Cambridge University Press, pp. 7–57.
Key Claims
- Census fertility forecasts are anchored to recent levels with r = +0.96 between the forecast's ultimate total fertility rate (TFR) and the 5-year pre-forecast average TFR; every turning point missed; 1996 Census TFR forecast of 2.245 is too high (no convergence assumed for race/ethnic fertility across immigrant generations).
- The Social Security Administration's (SSA) mortality projections since ~1980 project mortality decline at less than half the historical rate (.57%/year vs. 1.18%); discrepancy largest at younger ages; Lee-Carter projects e0 = 86.0 by 2070 vs. SSA's 81.5.
- The bundling pathology is quantified: Census bundles high fertility + low mortality → old-age dependency ratio (OADR) interval is less than a quarter the width of LT's; SSA bundles low fertility + low mortality → total dependency ratio (TDR) interval is near zero. Census assumes ρ(fertility, mortality) = −1.0; SSA assumes ρ = +1.0; LT observes ρ≈0.
- Random scenarios (Lutz et al.) improve on high-medium-low (HML) scenarios by avoiding the ±1.0 cross-correlation assumption, but retain perfect intertemporal correlation — fertility is always uniformly high or low in any given path, never allowing a baby boom midway through a medium trajectory. Cannot represent correct variance-covariance structure.
- Population aging is not a transitory baby boom effect: median LT OADR rises from 0.21 to 0.45 by 2072, and continues rising after the baby boom has died out — a permanent structural change.
- LT 95% probability intervals (PIs) for population growth rate are narrower than ex post Census forecast errors (which are 2–3× wider), showing stochastic intervals are not unreasonably wide.
- Government elderly spending rises from 8.5% → 22.5% of GDP by 2070; Old-Age, Survivors, and Disability Insurance (OASDI) accounts for only 29% of the increase; health care accounts for 57%. Fixing Social Security alone cannot solve the long-run fiscal problem.
- SSA's proposed 2% payroll tax hike (their estimated fix for 75-year actuarial balance) leaves a 75% probability of fund exhaustion before 2070 in stochastic simulation — policies that appear to work in the mean often fail in the majority of stochastic runs.
- Investing 90% of the trust fund in equities at 7% real return looks solvent in deterministic simulation but still leaves a two-thirds chance of exhaustion in stochastic simulation (median date shifts from 2031 to 2045).
- Raising the Normal Retirement Age to 71 by 2023: median actuarial balance > 0, but still 43% chance of exhaustion before 2070.
- Total federal, state, and local taxes rise from 24% → median 38% of GDP by 2070; 95% PI is 25%–53%.
Concepts Introduced or Extended
Entities Mentioned
Quotes
"The standard method for dealing with uncertainty in demographic (and many other) forecasts is the use of high, medium and low scenarios. This approach is deeply flawed, because it is based on very strong and implausible assumptions about the correlation of forecast errors over time, and between fertility and mortality."
"The [random scenario] approach does indeed seem preferable to the traditional scenario approach, since it avoids the false assumption that fertility and mortality forecasting errors are correlated either +1.0 or –1.0. However, it still assumes that errors in fertility (and mortality) are perfectly correlated over time."
"Fixing Social Security will not by itself solve the long run budgetary problems, although it obviously must be an important part of any solution."
My Take
The paper's most important contribution beyond the 1994 JASA paper is two-fold: the Lutz random-scenario critique (identifying intertemporal correlation as a second independent flaw) and the fiscal applications showing that standard policy fixes are stochastically much weaker than their deterministic justifications imply. The equity investment result is striking — what looks like a robust solvency fix deterministically is still a gamble in 2 out of 3 runs. The 57% health-cost share of elderly spending growth is also an early statement of what became a major fiscal policy theme. The main limitation is the exclusion of health care cost uncertainty from the stochastic framework (health cost per enrollee treated as deterministic), which the authors acknowledge sets a lower bound on true fiscal uncertainty.