Social Security claiming optimization — the modeling problem behind the calculator

WealthSchema StaffTax & retirement modelingMay 8, 20265 min read

The standard Social Security calculator takes birth year and projected benefit, runs an expected-value calc across claim ages 62, 67 (FRA), and 70, and returns the max-PV age. The math is one paragraph and the answer is usually "70."

That answer is correct for single-person, population-life-expectancy, no-other-tax-events households — a thin slice of the actual claiming-age population. Married households are joint optimization problems over two correlated mortality distributions. Households with documented health risk run under conditional life expectancy, not population baseline. Households doing Roth conversions in early retirement couple the claim decision to the conversion plan because both feed combined income. A real claiming engine handles all of these surfaces; the standard calculator handles one.

What the simple calculator gets right

The simple calculator gets the basic actuarial relationship right. Social Security benefits are reduced for claiming before Full Retirement Age (FRA, age 67 for those born 1960+) and increased for claiming after, by amounts that approximately offset on an actuarial-expected-value basis if you live exactly the actuarial-expected lifespan.

For a single individual claiming on their own record:

Formula

Single-person actuarial relationship

benefit(age) ≈ PIA × adjustment(age − FRA)
adjustment(−5) = 0.70  (claim at 62 if FRA=67)
adjustment(0) = 1.00   (claim at FRA)
adjustment(+3) = 1.24  (claim at 70 if FRA=67)

PIA: = Primary Insurance Amount — the base benefit at FRA
adjustment: = Multiplier on PIA based on early/late claim. Step function with monthly resolution.

The simple calculator computes PV of lifetime benefits at each claim age and picks the max. For population life expectancy, age 70 typically wins by ~$10K–$30K of expected lifetime PV. The recommendation is right for the population mean.

Where the simple calculator fails

The simple calculator misses several constraints that bind in real households.

Survivor benefits

When a spouse dies, the surviving spouse can step up to the higher of their own benefit or the deceased's benefit. This means the higher-earning spouse's claim age affects not just their own lifetime benefits but the surviving spouse's lifetime benefits — potentially decades after death.

For two-spouse households, the optimization is:

Formula

Two-spouse joint optimization

max E[Σ_t benefit(t) × P(any spouse alive at t)]

benefit(t): = Sum of spouse-1 benefit (if alive), spouse-2 benefit (if alive), or survivor benefit (if one alive)
P(any spouse alive at t): = Joint survival probability — uses the longer-lived spouse's life expectancy distribution

The optimization usually recommends the higher-earning spouse delay to 70 (maximizing the survivor benefit) and the lower-earning spouse claim at FRA or earlier (capturing more pre-mortality benefit). The single-person optimizer misses this by treating each spouse independently.

The optimizer's output is asymmetric: typically { higher_earner: claim_age=70, lower_earner: claim_age in [62, FRA] }. The "both at 70" answer a single-person calculator produces leaves expected joint PV on the table for most two-spouse households.

Spousal benefits

A spouse can claim a spousal benefit equal to 50% of the higher earner's PIA at FRA, instead of their own benefit, if the spousal benefit is higher. The lower-earning spouse usually has a spousal-benefit option that's larger than their own benefit. The interaction with claim timing:

The lower earner can only claim a spousal benefit if the higher earner has already filed.
The lower earner's spousal benefit is reduced for early claims (just like their own benefit).
"File and suspend" was a strategy that allowed the higher earner to file (enabling the spousal benefit) without actually receiving payments. It was eliminated for new claims after April 30, 2016. Calculators that still reference it are out of date.

Health-conditional life expectancy

Population life expectancy is the average across all 65-year-olds. Individual conditional life expectancy varies meaningfully by diagnosis, family history, and demographics. A household where one spouse has a documented life-shortening condition runs a different optimization than the population baseline. The engine's typical output flips for the affected spouse: claim early to capture pre-mortality benefits while the surviving healthy spouse's claim still funds the survivor benefit.

A real engine accepts the conditional distribution as a per-spouse input. The standard calculator running the population mean produces a recommendation that is correct for the average claimant and wrong for the specific household whenever the household's distribution materially differs.

Tax-aware claiming

Social Security benefits are partially taxable based on "combined income" (AGI + tax-exempt interest + half of SS). The thresholds:

	Filing status	Combined income range	Taxable portion of SS
Single	< $25,000 / $25K–$34K / > $34K	0% / up to 50% / up to 85%
MFJ	< $32,000 / $32K–$44K / > $44K	0% / up to 50% / up to 85%

Roth conversions, RMDs, and capital-gain harvests in the same tax year push combined income above the 85% threshold, dragging 85% of the SS benefit into taxable income. In that case the engine often delays SS past the actuarial-optimal age — the early-claim years would be taxed at marginal rates high enough to wipe the actuarial gain.

The claim decision and the other-income decisions are a joint optimization on combined income. An engine that runs SS in isolation from the conversion plan and the RMD schedule misses the dominant constraint for high-balance retirees.

The optimization problem, fully stated

A real Social Security claiming optimization for a married household:

Input 1
PIAs for both spouses
From SSA statements or estimated from earnings record.
Input 2
Conditional life expectancy distributions
Population baseline + health adjustments + family-history adjustments.
Input 3
Cash-flow needs in early retirement
Whether early SS is needed for living expenses or whether the household has other income sources.
Input 4
Tax projection
Other income (RMDs, conversions, capital gains) by year. Drives combined-income calculation.
Input 5
Other goals (charitable, estate, etc.)
Affect the relative valuation of late-life vs early-life cash flows.
Solver
Joint expected-PV maximization
Over claim ages for both spouses, joint over both life-expectancy distributions, with tax adjustment per year.
Output
Recommended claim ages plus sensitivity
Optimal pair of claim ages plus how the recommendation changes with life-expectancy and tax assumptions.

The optimization is small enough computationally to run in seconds. The hard part is the inputs — most users don't have ready answers for conditional life expectancy or accurate multi-year tax projections. Tools have to surface the inputs as dial-able parameters and let users see the recommendation conditional on their input.

What the engine returns across household profiles

The optimization output across the household profiles a claiming engine has to handle. None of these are recommendations to readers — they are the typical outputs an engine produces for a synthetic household with the given profile, used here to characterize the surface area an engine has to cover.

	Household profile	Typical engine output
Single, average health	Claim age 70 if cash flow allows; FRA if intermediate need; 62 only under cash-flow constraint.
Single, life-shortening diagnosis	Claim age 62 — maximum pre-mortality benefit.
Married, dual high earners, both healthy	Both claim age 70 under no cash-flow constraint. Otherwise higher earner 70, lower earner FRA.
Married, one high earner one low	Higher earner claim age 70 (drives survivor benefit). Lower earner FRA, or 62 under cash-flow need.
Married, age gap > 5 years	Output flips on relative life expectancy. Commonly: older spouse FRA, younger spouse 70.
Married, one spouse with life-shortening diagnosis	Affected spouse 62. Healthy spouse 70 when survivor benefit is the dominant lifetime PV term.
Pre-Medicare household drawing on ACA subsidies	Delay SS to preserve subsidies. Engine surfaces bridge-income source as a coupled constraint.
Early retiree with multi-year Roth conversion plan	Delay SS through the conversion window. Resume per the conversion schedule's final year.

The engine output rarely matches the simple-calculator recommendation. The shifts are driven by life expectancy, the combined-income tax surface, and cash-flow constraints — not by the gross-benefit actuarial math the simple calculator runs.

What this means for synthetic test data

Test data for SS optimization engines has to span:

Single individuals at every claiming age
Two-spouse households with various PIA ratios and age gaps
Households with each filing-status / combined-income / SS-taxability combination
Health-compromised individuals with shortened life expectancy
Pre-Medicare households with ACA subsidies in play
Households executing Roth conversion strategies
Multiple-record-of-earnings cases (DEC, divorced spouse with own earnings record)
Disability-onset cases (SSDI conversion to retirement)

Each scenario exercises a different optimizer code path. A test corpus of mid-career married couples with average earnings won't catch any of the meaningful claim-age failure modes.

Key takeaways

The standard 'delay to 70' recommendation is correct for single, average-health, no-other-income claimants — a thin slice of real claimants.
Two-spouse households are joint optimizations on two correlated mortality distributions. Survivor benefits dominate the higher earner's PV; pre-mortality benefits dominate the lower earner's.
File-and-suspend has not existed for new claims since April 30, 2016. Any engine still surfacing it is recommending a strategy that no longer exists.
Combined-income taxation couples SS timing to Roth conversions, RMD schedule, and ACA subsidy preservation. Engines running SS in isolation miss the dominant constraint for high-balance retirees.
Conditional life expectancy is the single input that flips the engine's output between 'claim at 62' and 'claim at 70'. Calculators that don't accept it have no path to the flip.
Test data for claiming engines needs cross-cutting scenarios: married vs single, baseline vs life-shortening health, with vs without other tax events. Each combination exercises a different code path.

Frequently asked questions

How does the optimization handle SS solvency risk?+

Cautiously. The SS Trust Fund is projected to be depleted in the 2030s without legislative action, after which benefits would automatically reduce to ~80% of scheduled. Most optimizers run the recommendation under both 'full benefits continue' and 'haircut to 80% post-depletion' scenarios. The optimal claim age is generally robust across both scenarios — the pre-depletion years are the same, and the haircut applies similarly to all claim ages — but the recommendation language should acknowledge the uncertainty rather than ignore it.

What about the divorced-spouse benefit?+

A divorced individual married 10+ years with no current marriage can claim against the ex-spouse's record (50% of ex-spouse's PIA at FRA, or up to 100% as survivor if ex has died). The ex doesn't have to know or consent. Optimizers serving this population have to track the divorced-spouse case explicitly; many products miss it. The interaction with the divorced individual's own record (claim either, but not both) requires careful logic.

Should we model conditional benefit increases (delayed retirement credits past 70)?+

Delayed retirement credits stop accruing at age 70. There is no benefit to claiming after age 70 — the household just loses unclaimed months. Optimizers should warn aggressively if a user enters a claim age past 70; this is a common error from users who confuse FRA and the maximum claim age.

How important are taxation differences between SS retirement, SSDI, and SS survivor benefits?+

Mostly the same income-tax treatment, but with one nuance: SSDI converts to SS retirement at FRA, and the retirement benefit is calculated using SSDI rules (effectively allowing earlier high-earnings years to count, which can raise the benefit). Optimizers serving disabled individuals approaching FRA should model the conversion explicitly.