Javier Meseguer

Overview

Researcher at the Office of Research, within SSA's Office of Research, Evaluation, and Statistics (ORES), Office of Retirement and Disability Policy, Social Security Administration. Works on disability insurance program analysis, mortality, and diagnostic trends.

Key Contributions / Features

• BVAR mortality forecasting (Meseguer 2006): SSA Office of Policy working paper. Proposed Bayesian Vector Autoregression (BVAR) with the Minnesota prior as an alternative to Lee-Carter for long-range US mortality forecasting (1928–2001 OCACT data, 21 age groups, combined sexes). Key contributions: (1) demonstrated that the Minnesota prior's shrinkage toward random-walk-with-drift resolves the overfitting problem for full VAR systems with m=21 age groups (462 parameters at p=1 exceeds the 74-year sample); (2) showed that calibrating cross-equation weights from the empirical cross-age correlation matrix [BVAR(1)-II] outperforms a univariate-equivalent specification [BVAR(1)-I] by 202 log-likelihood units; (3) demonstrated that BVAR credible intervals are 1.84×–3.0× wider than Lee-Carter's at a 75-year horizon because the Bayesian framework incorporates parameter uncertainty, while LC treats $\alpha_i$ and $\beta_i$ as known constants. The paper is the theoretical precursor to Meseguer (2008)'s empirical finding that LC intervals are catastrophically miscalibrated out-of-sample. See SSA Mortality Forecasting.
• Stochastic mortality forecasting comparison (Meseguer 2008): ORES Working Paper No. 111. Compared out-of-sample performance of bias-corrected Lee-Carter (LC) vs. AR(1) (Denton-Feaver-Spencer 2005) using HMD data for 16 countries and a 1980 jump-off year. Key findings: point forecasts nearly identical (RMSE ≈ 1); both models overestimate mortality at ages 65–95+ (97–99% of total MSE concentrated there); LC intervals catastrophically too narrow (0% $\delta_2$ coverage in 8/16 countries at 16+ year horizons); AR(1) intervals adequately calibrated; $e_0$ coverage is a misleading aggregate criterion (cancellation effect). Recommends AR(1) for OASDI policy use. See SSA Mortality Forecasting and Period Mortality.
• Outcome variation by diagnosis (Meseguer 2013): Using a Bayesian hierarchical multinomial logit model on 462,578 DI applications (1997–2004), showed that primary diagnosis accounts for ~44% of total variation in initial allowance decisions (ICC = 43.89%), while state of origin accounts for only ~6.2%. A diagnosis-only model outpredicts a full model including age, sex, earnings, employment status, and state. High positive correlation (r ≈ 0.56–0.74) between diagnosis-level initial and final allowance predictions indicates that the ranking of impairment severity is preserved across adjudicative stages. Diagnosis-level allowance rate spread: lung cancer ~94% initial allowance; degenerative back disorders ~23%. See SSA Sequential Determination Process and Listing of Impairments.
• Comorbidity patterns in DI/SSI diagnoses (Meseguer 2018): Using a high-dimensional Bayesian multivariate probit model on a 10% sample of 2009 DI claimants (n=157,835), estimated 5,356 pairwise correlations between the 100 most common diagnoses. Key findings: (1) comorbidity has risen sharply — 56.3% of 1997 claimants had a secondary diagnosis, rising to 71.4% by 2010; (2) mental disorders are systematically undercounted by primary codes alone — affective/mood disorders appear as secondary for 22.9% but primary for only 14.2% of claimants; (3) the mental-musculoskeletal comorbidity presumption is empirically wrong — affective/mood and back disorders are negatively correlated; (4) a neurological cluster of five diagnostic groups (organic mental disorders, TBI, brain neoplasms, late effects of cerebrovascular disease, nervous system disorders) shows strong positive inter-group correlations; (5) 52.5% of qualified 2009 DI claimants applied concurrently to SSI. See SSA Sequential Determination Process and Listing of Impairments.
• DI beneficiary mortality synthesis (Meseguer 2021): Synthesized nine SSA actuarial studies spanning DI mortality 1968–2015 into a single comparative framework. See DI Beneficiary Mortality.
• Cause-specific mortality by race/ethnicity (Meseguer 2024): Using CDC WONDER death records and ICD-10 classifications, documented age-adjusted cause-specific mortality trends 1999–2019 for four RE groups (WNH, Hispanic, Black, API) across 17 cause-of-death categories. Key findings: Black males overtook WNH males in poisoning mortality in 2019 (fentanyl third wave); WNH females had the worst relative poisoning increase of any female RE group (6.5×); post-2010 midlife mortality reversed for every RE/sex group except Black and API women; cancer disparities narrowed while cardiovascular improvement stalled; the Hispanic mortality paradox holds for all causes except digestive diseases, diabetes, and perinatal conditions. See Mortality by Race and Ethnicity.
• Full demographic system BVAR (Meseguer 2010): Unpublished SSA manuscript. Extends the mortality-only BVAR of Meseguer (2006) to a joint model of mortality, fertility, and population, projecting through 2100 using OCACT data. Two variants: Diff-BVAR (diffuse prior) and Inf-BVAR (informative prior anchored to SSA Alt 2 via the Villani 2006 mean-adjusted BVAR). Key findings: (1) Lee-Carter's 90% CI for female population aged 65+ does not cover the SSA Alt 2 projection from 2009 to 2088 — an 80-year interval non-overlap; (2) LC and Diff-BVAR produce nearly identical LE point forecasts but LC's aged dependency ratio exceeds Alt 2's by 3.3 pp (0.466 vs 0.433 in 2100); (3) Inf-BVAR tracks Alt 2 in expectation while providing appropriately wider uncertainty bounds than LC, making it the most policy-suitable specification. See SSA Mortality Forecasting and Lee-Carter Model.
• Business cycle applicant characteristics (Lindner, Burdick, and Meseguer 2017): Business cycle effects on DI applicant composition and the conditional applicant mechanism. See Conditional DI Applicants.

Related

• Social Security Administration
• Disability Insurance Program
• DI Beneficiary Mortality
• Mortality by Race and Ethnicity
• Conditional DI Applicants
• SSA Mortality Forecasting
• Period Mortality
• Stephan Lindner

Sources

• Meseguer 2006 — Long-Range Forecasts of Mortality and Life Expectancy Using Bayesian Vector Autoregressions
• Meseguer 2008 — The Out-of-Sample Performance of Stochastic Methods in Forecasting Age-Specific Mortality Rates
• Meseguer 2013 — Outcome Variation in the Social Security Disability Insurance Program The Role of Primary Diagnoses
• Meseguer 2018 — Correlation Patterns between Primary and Secondary Diagnosis Codes in the Social Security Disability Programs
• Meseguer 2021 — DI Beneficiary Mortality vs. General Population
• Meseguer 2024 — Trends in Cause-Specific Mortality by Race and Hispanic Origin, 1999–2019
• Meseguer 2010 — Forecasting Vital Rates and Population with Bayesian Vector Autoregressions
• Lindner Burdick and Meseguer 2017 — Characteristics and Employment of Applicants for Social Security Disability Insurance over the Business Cycle
• Engelman et al 2017 — Work Disability Among Native-born and Foreign-born Americans On Origins, Health, and Social Safety Nets
• Raut 2017 — Exits from the Disability Insurance Rolls Estimates from a Competing-Risks Model
• Strand and Trenkamp 2015 — When Impairments Cause a Change in Occupation