Overview
William R. Bell is a statistician at the U.S. Bureau of the Census (later U.S. Census Bureau). His methodological work spans stochastic forecasting of age-specific vital rates, principal-components time-series decomposition of demographic data, and comparative out-of-sample evaluation of forecasting methods. He is co-author of the Thompson-Bell-Long-Miller (1989) fertility forecasting framework — which directly inspired the Lee-Carter (1992) mortality model — and co-developer of the Bell-Monsell (1991) full-principal-component mortality forecasting approach.
Key Contributions / Features
- Bell (1997) comparative evaluation: Head-to-head out-of-sample MAPE comparison of HP, LC, Bell-Monsell, and RWD methods for age-specific fertility and mortality; central finding: LC/HP without bias adjustment lose to age-specific random walk 2–3×; LC with bias adjustment (jump-off fix) ≈ RWD (ratio 0.98–1.07); Bell-Monsell marginally worse than RWD without bias (ratio 1.03–1.22); firm conclusion: bias adjustment is non-negotiable for structured methods
- Bell-Monsell (1991): With Brian C. Monsell; applies SVD to mortality with all principal components retained and modeled as time series (vs. Lee-Carter's single-PC); captures more short-run age co-movement but does not systematically beat RWD
- Thompson et al. (1989): With Patrick A. Thompson, John F. Long, and Robert B. Miller; parameterized gamma-curve approach to age-specific fertility forecasting; reduced 32 age-specific fertility series to 3 parameters (TFR, MACB, SDACB) modeled as multivariate ARIMA; introduced the bias adjustment for short-term accuracy; provided the methodological template Lee and Carter (1992) adapted for mortality; used in the Census Bureau's 1988 national population projections
Related
Sources