Nonhealth Risk and DI Insurance Value

Definition

The value of disability insurance (DI) to recipients derives not only from the program's targeting of individuals with severe health conditions but also — and predominantly — from insurance against nonhealth financial risks. Deshpande and Lockwood (2022) show that among U.S. Social Security Disability Insurance (SSDI) recipients and eligible nonrecipients, the dominant source of the program's insurance surplus is within-health-category financial precarity: both less-severe and more-severe health groups in the applicant pool face elevated rates of mass layoffs, evictions, foreclosures, and bankruptcies before program entry, and this shared vulnerability accounts for $63\%$ of DI's insurance value. The remaining $37\%$ reflects between-health-category targeting — the standard assumption that health severity predicts financial need.

Key Ideas

The insurance value of DI is conventionally assumed to derive from health targeting: high-severity applicants have low earnings capacity and high financial need, so awarding them benefits produces a larger welfare gain than a cost-equivalent tax cut. Deshpande and Lockwood show this is quantitatively secondary.
Within each health-severity category, applicants face a wide distribution of financial circumstances. The fact that less-severe recipients and more-severe recipients have nearly identical pre-entry consumption ( $\approx \$17{,}000$ vs. $\approx \$16{,}400$ ) and similarly high adverse event rates ( $\approx 57\%$ vs. $\approx 50\%$ had at least one mass layoff, eviction, foreclosure, or bankruptcy in the prior four years) means the program is providing substantial insurance against shocks that are not predicted by health severity alone.
The fact that more-severe nonrecipients (M-NDI) are substantially better off than either recipient group — $10$ – $30\%$ fewer adverse events, higher earnings, more educated and married — shows that health severity is not a reliable proxy for financial need within the applicant pool.
Surplus per recipient: All U.S. Disability Program (USDP) $\$8{,}700$ ( $64\%$ above a cost-equivalent tax cut); M-DI (more-severe recipients) $\$9{,}900$ ; L-DI (less-severe recipients) $\$7{,}700$ ; M-NDI $-\$2{,}200$ .
USDP generates more surplus than a hypothetical program with perfect health targeting ( $\$8{,}700$ vs. $\$5{,}000$ ), because perfect targeting would eliminate less-severe recipients who are genuinely financially vulnerable.
Marginal Value of Public Funds (MVPF) for USDP reforms $\approx 1.42$ , compared to unemployment insurance (UI) $\approx 0.61$ .

Empirical Foundation: The Financial Distress Facts (Deshpande, Gross, and Su 2021)

Before Deshpande and Lockwood (2022) could decompose DI's insurance value, the adverse event rates that drive the Expected Additional Willingness to Pay (EAWTP) calculation had to be measured. Deshpande, Gross, and Su (2021) built the administrative data infrastructure — linking the universe of Social Security Administration (SSA) 831 applicants (2000–2014) to nationwide bankruptcy records (Gross et al. 2016), CoreLogic foreclosure records, and AIRS eviction records — and established three facts:

Applicants face high absolute rates of financial distress: conditional on homeownership, disability applicants' foreclosure rate is $\approx 2\times$ the general population rate; $12\%$ ever file for bankruptcy; $18\%$ ever experience eviction (among non-homeowners).
"Ashenfelter's peak": Financial distress peaks at the disability application date relative to both the general population and the applicants' own lifetime profile. This means applicants are in the highest-marginal-utility state precisely when they apply — the consumption-smoothing value of DI is largest at the moment of entry.
Disability allowance causally reduces distress: Vocational grid regression discontinuity (RD) (ages 50, 55) yields instrumental variables (IV) estimates of $-31\%$ bankruptcy, $-34\%$ foreclosure, $-15\%$ net home sale, $+14\%$ home purchase within 3 years of initial allowance. The effect is primarily a wealth effect from cash transfer + health insurance.

These three facts are the empirical inputs to Deshpande-Lockwood (2022)'s EAWTP framework: the high adverse event rates (facts 1–2) establish that applicants draw on DI from a position of financial precarity; the causal estimates (fact 3) confirm that the program delivers consumption-smoothing specifically in high-marginal-utility states. Together they imply that a large share of DI's insurance value reflects within-health-group financial vulnerability — the $63\%$ nonhealth share.

The EAWTP Framework

Deshpande and Lockwood (2022) adopt a sufficient-statistics approach. The EAWTP for DI is defined as the dollar amount the average recipient would pay for program access above its actuarial cost. Formally:

EAWTP = Program cost × marginal utility markup

The marginal utility markup is the ratio of the recipient's expected marginal utility of consumption to the counterfactual (no-DI) marginal utility — capturing how much the program shifts consumption toward states of high marginal utility. The decomposition:

Health risk channel: Recipients with more severe health have lower counterfactual earnings capacity $\to$ higher counterfactual financial need $\to$ higher marginal utility. Targeting them produces a between-health-group insurance gain.
Nonhealth risk channel: Within each health severity group, recipients are drawn from the bottom of the financial circumstances distribution (high adverse event rates, low earnings). This within-group concentration produces an insurance gain even holding health severity constant.

The empirical finding is that channel 2 accounts for $63\%$ of total insurance value. Robustness: using a 30-category health classification still yields a nonhealth share $>50\%$ .

Welfare weights follow Hendren (2020): government spending is evaluated using inverse-optimum weights calibrated to the observed income tax schedule, ensuring that $\$1$ to a lower-income individual receives higher weight.

Selective Application: The Mechanism

The paper's key structural finding distinguishes between two potential sources of program value:

Selective awards: SSA's adjudication process screens applicants and awards benefits only to those meeting medical and vocational criteria.
Selective application: The Substantial Gainful Activity (SGA) earnings limit ( $\$1{,}000$ – $\$1{,}500$ /month) screens workers before they ever apply — individuals earning above SGA cannot qualify and so self-select out of the applicant pool.

Counterfactual experiment — Random-App-DI: Randomly assigning DI to all individuals across all four groups (L-DI, M-DI, M-NDI, and less-severe nonrecipients L-NDI) produces a surplus of $-\$6{,}100$ — strongly negative. This means the general population of DI-eligible individuals would not benefit from receiving DI. The program only generates positive value because the subset who actually apply is drawn from the financially vulnerable lower tail.

Counterfactual experiment — Earnings-Test-DI: A hypothetical program that retains the SGA earnings limit (selecting on work incapacity) but eliminates all medical screening generates a $\$5{,}900$ surplus — positive and recovering $68\%$ of USDP's $\$8{,}700$ . The SGA limit alone, without any health screening, produces most of USDP's insurance value. This is the mechanism.

The implication inverts the standard narrative: SSA's elaborate medical determination process — the five-step sequential evaluation, Listings, Residual Functional Capacity (RFC) assessments, Administrative Law Judge (ALJ) hearings — adds only $32\%$ of program value beyond what simple work-incapacity screening achieves.

Contrasting Screens: SGA vs. Field Office Access

Deshpande and Lockwood (2022) clarify why the SGA earnings limit is a beneficial screen while field office closings (Deshpande and Li 2019) are a harmful screen:

Feature	SGA Earnings Limit	Field Office Closings
Operates on	Work capacity: screens workers with viable earnings alternatives	Administrative friction: screens applicants unable to navigate bureaucracy
Who is deterred	Higher-earning workers with meaningful outside options	Medium-severity, high-need applicants deterred by congestion
Effect on targeting	Improves: applicant pool is more financially vulnerable	Worsens: applicant pool includes more denied-quality, lower-need cases
Welfare effect	Positive	Negative (net social cost $\approx \$1.2$ billion)

The key distinction: SGA screens on the dimension that predicts financial need (earnings capacity), while distance and congestion screen on the dimension that predicts bureaucratic navigation ability, which is uncorrelated with financial need. See DI Application Costs and Take-Up.

Why More-Severe Nonrecipients (M-NDI) Don't Benefit

M-NDI are the group whose non-participation is not a problem. They are more-severely impaired individuals who do not apply for or receive DI. The paper finds they:

Have $10$ – $30\%$ fewer adverse nonhealth events than either recipient group
Are more likely to be married (providing household insurance)
Have higher educational attainment
Have counterfactual earnings of $\approx \$57{,}500$ /year (vs. $\$6{,}600$ for L-DI and $\$3{,}600$ for M-DI)

This suggests M-NDI have chosen not to apply because they have adequate outside resources — household earnings, private insurance, or savings. Providing them DI would produce a negative surplus of $-\$2{,}200$ /recipient: they don't value the benefit enough to justify its cost because their counterfactual consumption loss from disability is small relative to program cost. This finding undermines the view that the $\approx 55\%$ of more-severe individuals who do not receive DI represent a policy failure. See DI Denied Population.

Implications for the Causal Effects Literature

The paper complements the causal effects literature (Maestas, Mullen, and Strand (MMS) 2013; French and Song 2014) but addresses a different question. MMS and French/Song estimate the causal work-disincentive of DI receipt — how much DI reduces labor supply for marginal recipients. Deshpande and Lockwood ask whether the resulting transfer produces enough insurance value to justify this efficiency cost.

The finding that marginal applicants' counterfactual earnings are only $\$6{,}000$ – $\$9{,}000$ /year means that even the $\approx 28$ pp labor force participation (LFP) reduction documented by MMS translates into a small absolute earnings loss — approximately $\$1{,}680$ – $\$2{,}520$ /year per marginal applicant displaced from work. Against a program benefit of $\approx \$10{,}000$ – $\$15{,}000$ /year, the efficiency cost is modest relative to the insurance gain. See Causal Effects of DI Receipt.

Why It Matters

For DI Program Design

The paper establishes that the SGA earnings limit is a load-bearing element of DI's insurance value — not merely a work-test formality. Reforms that weaken the SGA limit (e.g., by raising it substantially or eliminating it) risk admitting higher-earning individuals who would reduce the program's average insurance value. Conversely, reforms that preserve the SGA limit while streamlining the medical determination process could maintain most of DI's value at lower administrative cost.

For the MVPF Literature

DI's MVPF of $\approx 1.42$ is substantially higher than UI's $\approx 0.61$ . This ranking reflects the more selective applicant pool and stronger self-selection mechanism in DI. The comparison suggests that marginal expansions of DI (e.g., lowering the evidentiary threshold for severe impairments) generate more social value per dollar than marginal expansions of UI.

For the Optimal Benefit Level (Chetty 2006 and Meyer-Mok 2013)

The EAWTP framework asks whether DI generates positive surplus at the margin of program entry — but a separate question is whether the benefit level itself is optimal. Meyer and Mok (2013) provide the empirical input to the Chetty (2006) general optimal benefit formula:

Chetty (2006) formula: The optimal benefit level satisfies $\gamma \cdot (\Delta c/c) \cdot [1 + \tfrac{1}{2}\rho \cdot (\Delta c/c)] \approx \varepsilon_{D,b}$ , where $\gamma$ is the coefficient of relative risk aversion, $\Delta c/c$ is the consumption drop from disability, and $\varepsilon_{D,b}$ is the elasticity of program take-up with respect to the benefit level.

Meyer-Mok estimate $\Delta c/c = -0.25$ for Chronic-Severe disability (food + housing consumption, 10 years post-onset), with an application rate $D = 0.12$ . Under this calibration:

For $\gamma = 3$ : the left-hand side $\approx 0.66$ , meaning DI is optimal only if $\varepsilon_{D,b} \geq 0.66$
The literature median estimate of $\varepsilon_{D,b}$ is $\approx 0.49$ (well below this threshold)
For $\gamma \geq 3$ and $\varepsilon_{D,b}$ near the literature median, current DI benefit levels appear sub-optimal — they undercompensate the severely disabled

The Deshpande-Lockwood (2022) surplus analysis and the Chetty-Meyer-Mok optimality analysis are complementary but address different margins: the former asks whether DI generates more surplus than a cost-equivalent tax cut (it does, by $64\%$ ); the latter asks whether the benefit level maximizes welfare conditional on program existence (it likely does not). Both findings are needed for a complete welfare assessment of DI.

For the Nonhealth vs. Health Risk Debate

The $63/37$ nonhealth/health decomposition challenges the design logic of targeting social insurance programs on categorical health conditions. If most of DI's value comes from insuring against job loss, credit shocks, and housing instability, then program reforms focused exclusively on improving health measurement — better imaging, tighter Listings criteria, more frequent Continuing Disability Reviews (CDRs) — address only $37\%$ of the program's value-generating mechanism.

For Household Insurance (Spousal Labor Supply)

DI provides an additional layer of insurance not captured in individual-level welfare analysis: it insures spouses against being forced into unwanted labor market entry. When a primary earner becomes disabled, standard household models predict the secondary earner will increase labor supply — the Added Worker Effect (AWE). Chen (2012) estimates DI income crowds out this response by $-6\%$ LFP (growing to $-8.4\%$ by year $5$ ). By replacing the lost household income, DI insures the spouse against an involuntary expansion of market work. This spousal insurance dimension is not accounted for in the Deshpande-Lockwood (2022) EAWTP framework, which focuses on recipients and eligible nonrecipients — suggesting the $64\%$ surplus estimate above a cost-equivalent tax cut is a lower bound on DI's true household-level insurance value. See Added Worker Effect.

Marital Status Heterogeneity in DI's Insurance Value (Autor, Kostøl, and Mogstad 2015)

Autor, Kostøl, and Mogstad (2015) directly quantify DI's household-level insurance value using Norwegian judge-lottery IV ( $N = 14{,}077$ , 1994–2005) combined with consumption imputed from administrative registers. The household-level perspective reveals a sharp heterogeneity that is invisible to individual-level analysis:

Fiscal cost accounting: Net fiscal cost per DI allowance is $\approx \$16{,}000$ /year after subtracting $\approx 30$ cents of benefit substitution per $\$1$ DI (other Norwegian disability/sickness transfers are partially crowded out), plus the labor supply cost.

Married applicants have near-zero DI insurance value: Household income change from denial: $-\$1{,}615$ /capita (n.s.); consumption change: $-\$0.83$ /capita (n.s.); benefit-to-cost ratio: $-0.37$ . The spousal AWE fully offsets the consumption loss — married households are self-insured at the household level. Willingness-to-pay (WTP) (static) $\approx \$2{,}700$ /capita.

Single applicants have high DI insurance value: Household income from allowance vs. denial: $+\$9{,}086$ /capita $(p < 0.01)$ ; consumption: $+\$9{,}835$ /capita $(p < 0.1)$ ; benefit-to-cost ratio: $0.86$ . With no spouse to supply the AWE, single applicants depend entirely on DI for consumption smoothing. WTP (static) $\approx \$9{,}100$ /capita — approximately $3.4\times$ the married rate.

This marital status heterogeneity enriches the Deshpande-Lockwood (2022) framework in two directions. First, it suggests the nonhealth risk share of DI's value may be concentrated among single applicants who face financial precarity without household backstops. Second, it implies that blanket DI eligibility reforms are less efficient than targeting: relaxing eligibility for single applicants would generate substantially more welfare per dollar of fiscal cost than relaxing it for married applicants. See Added Worker Effect, Andreas Ravndal Kostøl, and Magne Mogstad.

Open Questions

How does the nonhealth risk share vary across DI's diagnostic subgroups? Applicants with musculoskeletal disorders may have different nonhealth risk profiles than those with mental conditions.
Does the $63\%$ nonhealth share generalize to Supplemental Security Income (SSI), which serves a lower-income population with more severe financial precarity?
How does the MVPF comparison between DI and UI change if the UI recipient pool has become more financially vulnerable post-2008?
Would improving job-retraining or partial-disability programs reduce the nonhealth risk component by offering more workers viable alternatives to full DI application?