Summary
Michaud and Wiczer build a quantitative overlapping-generations (OLG) model in which Social Security Disability Insurance (SSDI) application is an "option" exercised by long-term non-employed workers facing macroeconomic and health risks. The model's key mechanism is that physically demanding occupations bundle two correlated hazards: accelerated work-limitation onset with age and secular wage decline — so that workers who most need the insurance option are also the ones for whom it becomes most attractive over time. A Shapley-Owen decomposition of simulated SSDI awards finds that pre-2000 secular wage declines account for 24% of new awards; post-2000 baby-boom aging accounts for 13%; and business cycle fluctuations contribute approximately zero to awards throughout, despite having large effects on applications. The model also documents an SSDI non-employment shadow — 1.6–2.5% of working-age population in the application pipeline at any time — which implies that empirical employment estimates omitting applicants understate SSDI's labor market footprint by 30–40%.
Key Claims
- The disability option: SSDI application requires 5 months of non-employment before filing; it is therefore an option on a future benefit stream, exercised by workers who exhaust labor market alternatives. This creates the SSDI "non-employment shadow."
- Occupational bundling: A 5 percentage point (pp) higher work-limitation hazard by age 60 is associated with a 1% lower secular wage trend in the same occupation. Physical occupations thus face double exposure — more likely to become disabled and more likely to find the wage-benefit tradeoff favoring SSDI.
- Shapley-Owen decomposition of the pre-2000 SSDI rise: secular wage declines = 24%, baby-boom demographics = 13%, business cycles ≈ 0%.
- Post-2000 period: baby-boom aging dominates (13%), wage declines smaller contribution, cycles ≈ 0% for awards throughout.
- Business cycles and applications: Business cycles have large effects on applications (elasticity ~0.07–0.17 to job loss) but near-zero effects on awards — consistent with compositional shifts among applicants, not eligibility standard changes.
- Award elasticities (Table 4): Application elasticity to secular wage decline = −0.20 (full population), −1.16 (d>0 health-impaired), −0.19 (d=0 healthy). Job loss elasticity ≈ 0.07–0.17. Unemployment rate elasticity ≈ 0 for awards.
- Non-employment shadow: 1.6–2.5% of working-age population are in the SSDI application pipeline; omitting them understates the measured employment effect of SSDI by 30–40%. Non-employed applicants grew from 25% of non-employed (1985) to 66% (2010).
- Male non-employment effect: Model predicts a 4.3 pp increase in male non-employment attributable to SSDI (vs. 6.4 pp actual 1985–2010); 75% is from current beneficiaries, 18% from applicants.
- Vocational awards puzzle: Holding vocational rules constant, the model can explain only ~6 pp of the 35 pp rise in vocational award share (from 25% to ~60% post-2010). The gap implies substantial de facto implementation drift at the vocational stage — consistent with structural evidence for judge leniency variation in French-Song (2014) and Deshpande et al. (2025).
- Welfare: Average SSDI beneficiary gains ~1% consumption equivalent; right-tail gains are large. Health-impaired workers (d>0) value SSDI 80–125% more than healthy workers (d=0). Both health groups respond similarly to job loss, but d>0 workers are far more responsive to wage trends — confirming wage trends, not job loss, as the health-gradient mechanism.
- Calibration: Occupational Information Network (O*NET) principal component analysis (PCA) (19 physical task measures → first principal component [PC]; 101 other knowledge, skills, and abilities [KSA] → first PC); Panel Study of Income Dynamics (PSID) for health transitions and wages with Heckman selection; Current Population Survey (CPS) for job flow hazards. Preference parameters from Low and Pistaferri (2015): θ=−0.448 (disability income utility shifter), η=−0.185 (health utility), γ=1.5 (constant relative risk aversion, CRRA). SSDI acceptance probability from Lahiri-Vaughan-Wixon (1995) marginal effects.
Concepts Introduced or Extended
- DI Growth Decomposition — adds structural Shapley-Owen decomposition: wage trends dominant pre-2000 (24%), demographics post-2000 (13%), cycles ≈ 0 for awards
- Conditional DI Applicants — formalizes the SSDI non-employment shadow; 1.6–2.5% of working-age population in pipeline; employment effect understated 30–40% by omission
- Vocational Grid — quantifies the vocational awards puzzle: constant rules predict ~6 pp of 35 pp actual rise; implies de facto implementation drift
Entities Mentioned
Quotes
"We find that when the economy is booming and when it goes bust, it matters little for awards. In contrast, when wages fall or the population ages, disabled workers are far more likely to use disability insurance."
"The SSDI non-employment shadow is growing: among non-employed workers, the fraction applying to SSDI rose from 25% in 1985 to 66% in 2010."
"Holding the vocational rules constant… the model can only account for a 6 pp increase in the vocational award share while the data show an increase of about 35 pp."
My Take
The paper's greatest contribution is its decomposition: the finding that business cycles matter for applications but almost nothing for awards is a sharp quantitative result that reframes debates about SSDI as a countercyclical stabilizer. The occupational bundling mechanism is novel and well-identified via O*NET PCA. The non-employment shadow (1.6–2.5% in pipeline) is an underappreciated measurement correction with real consequences for employment statistics. The vocational awards puzzle is the paper's most uncomfortable result for policy: if constant rules can explain only ~6 pp of a 35 pp rise, either the vocational rules have drifted substantially in implementation or there has been a large unobserved health deterioration among the relevant demographic — and the paper's own health-transition evidence argues against the latter. The paper predates Deshpande et al. (2025), which confirms the award-rate decline post-2010 is largely compositional, consistent with but not identical to the Michaud-Wiczer implementation-drift interpretation.