Policy 08: Eliminate HSA, FSA, and 401k.

Researched: 2026-07-22. Status: draft.

The plank

Each of these policies should be disconnected from employment. Raise limits on IRAs to match existing 401k limit.

README.md and app/templates/index.html carry identical wording for this plank — no drift. (Noted in passing while checking: Policy 4's title does drift between the two files — "Disintermediate healthcare and employment" vs "Separate healthcare and employment" — that belongs to the Policy 4 doc, not this one.)

Reading the plank precisely: it does not propose ending tax-advantaged saving. It proposes ending the employer-gated versions of it (401k, FSA, and the HDHP-tethered HSA) and moving the entire tax preference to the individually owned IRA, with the IRA limit raised to the current 401k employee-deferral limit. This doc researches that reading.

The problem

Three tax shelters — the 401k, the health FSA, and the HSA — are gated by your employer. Whether you get them, what they cost you in fees, and whether you forfeit money in them depends on where you work, not on anything you did.

The access gap. Retirement benefits were available to 72% of private-industry workers in March 2025, and only 53% participated — a 19-point gap (BLS, National Compensation Survey, September 2025). Access is 59% at establishments with under 100 workers vs. 90% at establishments with 500+ (BLS, March 2025). AARP counts roughly 57 million private-sector workers ages 18–64 — nearly half — with no way to save for retirement at work, and 81% of them (46 million) earn $50,000 or less (AARP Public Policy Institute, payroll-deduction savings research, 2022–2025).

The limit inequity. In 2026 the 401k employee deferral limit is $24,500 while the IRA limit is $7,500 (IRS, November 2025). Two workers with identical incomes get tax shelter differing by a factor of 3.3 purely because one employer sponsors a plan and the other doesn't. Counting employer contributions, the gap is worse: the total defined-contribution limit (IRC §415(c)) is $72,000 for 2026 (IRS Notice 2025-67; figure per the same IRS announcement cycle — verify against the notice text).

The fee drag. Ayres and Curtis, studying 3,500+ plans holding $120B+ (2010 data), found fees and menu restrictions cost the average 401k participant 78 basis points a year in excess of index-fund costs; 52% of plans offered at least one "dominated" fund; and in 16% of plans, fees fully consumed the tax benefit of the 401k for a young worker (Ayres & Curtis, "Beyond Diversification," Yale Law Journal 124:1346, 2015). Fees have fallen since, but the small-plan penalty persists: BrightScope/ICI found median total plan cost of 1.13% of assets for plans with $1M–$10M vs. 0.27% for plans over $1B (2014 data), and an average total plan cost of 0.81% in 2021 (BrightScope/ICI Defined Contribution Plan Profile, 2024 edition). The people stuck in the expensive plans are, systematically, employees of small firms — the same population with the worst access. An IRA at any major brokerage holds an index fund at 0.03–0.20%.

The forfeiture machine. The health FSA's use-it-or-lose-it rule transferred an estimated $4.5 billion from workers to employers in 2023 ($5.1B in 2022, when 52% of FSA holders forfeited something); EBRI puts the average forfeiture around $441 (Money/EBRI analyses, 2023–2025). Forfeited funds revert to the employer, and no agency tracks what happens to them.

The job-change tax. 41.4% of workers cash out their 401k at job separation, and nearly 90% of those drain the entire balance — paying income tax plus the 10% penalty (Yin et al., "Cashing Out Retirement Savings at Job Separation," Marketing Science, 2023; sample of 162,360 workers at 28 firms). Strikingly, leakage rises with the employer-match share of the balance — the "free money" framing nudges people to treat it as a windfall.

The HSA skew. HSAs are nominally portable but require HDHP coverage, which is chosen by employers for most insured workers (HDHP/SOs were 33% of covered-worker enrollment in 2025 — KFF Employer Health Benefits Survey, 2025). Households with incomes over $100,000 account for roughly 70% of HSA contributions (CBPP summarizing GAO, 2008; GAO-25-107480, 2025 finds the skew persists: in 2021, ~10% of returns under $75k AGI contributed vs. ~20% of returns with $500k–$1M AGI).

What the status quo costs. JCT scores the defined-contribution tax expenditure at $251.4 billion in 2024 (cash-flow basis) and roughly $1.4 trillion over FY2024–28 (JCT, Estimates of Federal Tax Expenditures FY2024–2028, JCX-48-24, December 2024); HSAs cost ~$11B in 2023 and ~$62B over 2022–26 (JCT, via Tax Policy Center, 2025). And the benefit is regressive: 58% of the retirement tax expenditure ($160B in 2019) went to the top-earning 20% of workers; 1% ($4B) went to the bottom 20%, over 80% of whom get nothing at all (Bipartisan Policy Center, "Who Benefits from Retirement Tax Breaks?", 2023, using 2019 data; consistent with CBO, Distribution of Major Tax Expenditures, 2021).

How it lands on an ordinary person: change jobs and you face rollover paperwork or a cash-out temptation that 4 in 10 people fail; misjudge December medical spending and your employer keeps the FSA balance; work for a small firm and your fund menu quietly skims an extra ~0.9%/year; work for one of the ~half of small employers with no plan and you get a $7,500 shelter while the person across the street gets $24,500 plus a match.

Who profits from the status quo

• Recordkeepers and asset managers. The DC industry administers nearly $13–14 trillion for ~140 million participants (PLANSPONSOR 2025 Recordkeeping Survey). Fidelity alone administers over $2.04 trillion of 401k assets for 31.7 million participants; Empower serves 12+ million; Vanguard, Alight, and Voya round out the top five (industry rankings, 2025). Back-of-envelope: even 20–50 bp of all-in revenue on $10.1T of 401k assets is $20–50 billion a year in administration and asset-management revenue tied to the employer channel (estimate, ours). At Ayres–Curtis's 78 bp excess (2010 data — fees have since fallen, so treat as an upper bound), the excess alone would be tens of billions annually.
• Employers, who keep ~$4.5B/year in FSA forfeitures (2023) and use vesting schedules and matches as retention handcuffs.
• HSA custodians, who sit on $174 billion across 41.7 million accounts (year-end 2025) collecting custodial and investment fees (Devenir 2025 Year-End HSA Research Report).
• The lobby. The Investment Company Institute spent $4.0 million on federal lobbying in Jan–Sep 2025 (OpenSecrets); the American Benefits Council ~$950,000 in 2024, on ERISA, pooled employer plans, and HSA rules (OpenSecrets); the securities & investment industry as a whole spent $195 million in 2025, up 26% (OpenSecrets, January 2026). These groups' opposition to the Biggs–Munnell proposal (below) is on the record.

To be fair to the villains: recordkeeping is a real service, fees have declined for two decades, and Fidelity/Vanguard also offer the cheap IRAs this plank would move money into. The factual claim is narrower — the employer channel adds an intermediation layer whose costs fall hardest on small-firm workers, and the industry lobbies to preserve the channel.

The proposal, concretely

What changes in law:

1. Repeal IRC §401(k) (cash-or-deferred arrangements) prospectively: no new employee deferrals after a transition date. Existing balances roll to IRAs (mechanics below).
2. Repeal the health FSA (IRC §125 cafeteria-plan health flexible spending arrangements). Use-it-or-lose-it ends with it. (Dependent-care FSAs need a separate answer — open question below.)
3. Repeal IRC §223 (HSA) for new contributions; existing HSAs either grandfather or merge into IRAs with a qualified-medical-withdrawal rule.
4. Raise the IRA limit (IRC §219/§408) to the 401k deferral limit — $24,500 at 2026 levels, indexed, with catch-ups aligned ($8,000 age-50 catch-up in 2026 vs. the current IRA catch-up of $1,100).
5. Universal payroll auto-IRA. This is the load-bearing mechanism. Every employer without a plan becomes a payroll conduit: default enrollment of every worker into a personal IRA at a default rate, invested in a default target-date fund, opt-out allowed. This is not speculative — it is the state auto-IRA model already operating: 17 states enacted, 15 operational, $2.75B+ saved by early 2026 (ASPPA, March 2026). OregonSaves (launched 2017) and CalSavers (2019) are the working prototypes. The federal Saver's Match arriving in 2027 (SECURE 2.0) pays into IRAs and stacks cleanly on this design (Pew, September 2025).
6. Employer contributions can continue as direct deposits to employee IRAs (SEP-style) or convert to wages; either way the money is the worker's from day one — no vesting schedules.
7. Transition for existing balances: 401k plans held $10.1 trillion at year-end 2025 (ICI Quarterly Retirement Market Data, Q4 2025, March 2026). The rollover rails already move enormous volume — $670B rolled from employer plans to IRAs in 2022 (ICI), with ~$941B projected for 2026 (Cerulli, via InvestmentNews). A phased wind-down (e.g., 5 years, default in-kind rollover to a low-cost default fund at the incumbent custodian, or to a TSP-style public option) is an order-of-magnitude scale-up of existing flows, not a new invention.

Where the model is already in use: the state auto-IRAs above; and structurally, the IRA itself already holds $19.2 trillion — more than 401k plans — and 39% of all US retirement assets (ICI, year-end 2025), with rollovers making up ~80% of average traditional-IRA balances (ICI/PLANSPONSOR, 2025). The IRA is not an experiment; it is already the terminal destination for most 401k money. The plank shortens the pipeline.

Evidence

• Payroll defaults drive saving; the employer plan per se does not. Madrian & Shea showed auto-enrollment massively raises participation (QJE 116(4):1149–1187, 2001); Thaler & Benartzi's Save More Tomorrow raised participant saving rates from 3.5% to 13.6% over 40 months (JPE 112(1):S164–S187, 2004); Vanguard reports 94% participation under auto-enrollment vs. 64% voluntary in 2025 (How America Saves 2025). Crucially, the active ingredients — payroll deduction, default enrollment, default fund — are all portable to IRAs.
• State auto-IRAs prove the IRA channel works. CalSavers: ~$1.55B and ~590,000 funded accounts in late 2025, ~$1.70B and 736,421 contributing accounts by February 2026; OregonSaves: $445M and 190,877 contributing accounts at December 31, 2025 (ASPPA/PSCA program trackers, 2025–2026). AARP: workers with payroll deduction are 15x more likely to save than without (AARP PPI, 2022). Caveats: these programs are young, average balances are small (~$2,300–$2,600 implied by the figures above), default rates (typically 5%) sit below the ~12.1% combined employee+employer rate in 401k plans (Vanguard, 2025), and opt-out rates run higher than in employer plans.
• The 401k's measured advantage is smaller than its reputation. Choi, Laibson, Beshears et al. find auto-enrollment raises steady-state saving by only ~0.6% of income once job turnover, cash-out leakage, and auto-escalation opt-outs are counted (NBER Working Paper 32828, August 2024). The gap between lab-famous effects and real accumulation is largely because the account is tied to jobs people leave.
• The tax preference buys little new saving. Biggs and Munnell argue the DC tax expenditure mostly rewards saving that would happen anyway and accrues to high earners (CRR, "The Case for Using Subsidies for Retirement Plans to Fix Social Security," January 2024). This plank redeploys rather than repeals the preference, but the finding undercuts the claim that trimming employer-channel privileges will crater national saving.
• Fees: Ayres & Curtis (2015) and BrightScope/ICI (2024 edition), as above.

Overall strength of evidence: moderate. The problem side (coverage gap, fees, forfeitures, regressivity, leakage) is strongly documented with government and peer-reviewed sources. The solution side rests on state auto-IRA programs that are real but young and small relative to the $10T system they would replace. No country has executed a wholesale employer-DC-to-individual-account migration; this would be a first.

Who wins, who loses

Does compensation compensate? The plank's implicit compensation for match-losers is that matches are deferred wages and competitive labor markets should convert them to cash or IRA deposits. Economic theory supports this; clean empirical evidence on benefit-to-wage pass-through is mixed and slow (estimate, weak evidence — flagged). For high savers, raising the IRA limit to $24,500 compensates fully for lost employee deferrals but not for lost employer contributions above that; that loss is real and intended — it is the regressive slice of the current expenditure. Industry job losses would phase over the wind-down period (5+ years) and partially convert to IRA-side employment at the same firms.

Fiscal impact

No CBO/JCT score exists for this exact plank; what follows combines official scores of the pieces with labeled back-of-envelope arithmetic.

• Baseline costs: DC-plan tax expenditure $251.4B in 2024 cash-flow, ~$1.4T over FY2024–28 (JCT, JCX-48-24). HSA: ~$11B in 2023, $62B over 2022–26 (JCT). Health-FSA salary reductions escape both income and payroll tax inside the larger employer-health exclusion (no separate JCT line commonly published — unverified as a standalone number).
• The swap is roughly revenue-neutral on employee money. Raising the IRA limit to the 401k limit while abolishing 401k deferrals largely relabels the same deferral. Net revenue effects come from the pieces that don't carry over:
• Employer contributions above what converts into (taxable-then-IRA-deferred) wages become taxable compensation. Full repeal of the DC preference — much more aggressive than this plank — was estimated by Biggs–Munnell at +$121B/yr (2024); this plank captures only a fraction, plausibly $20–50B/yr (back-of-envelope, ours).
• HSA repeal: ~+$14B/yr by late-2020s trajectory (JCT-based extrapolation, ours).
• FSA repeal: single-digit billions/yr of income+payroll revenue (back-of-envelope, ours); partially offset because FSA forfeitures currently return to employers as taxable-ish plan offsets.
• Payroll-tax interaction: employer retirement contributions currently escape FICA; wage conversion brings them in — an additional revenue gain and a Social Security solvency assist (this is the core of the Biggs–Munnell argument).
• Honest caution: cash-flow tax-expenditure numbers overstate the long-run cost of deferral (tax is paid later, not never); ICI's Peter Brady has pressed this critique for years ("The Tax Benefits and Revenue Costs of Tax Deferral," ICI, 2012 — unverified in this pass). Any claim that this plank "raises $250B/yr" would be wrong; the defensible claim is net positive, likely tens of billions per year, from the employer-side exclusions and HSA/FSA repeal, plus administrative savings to households.

The fiscal case is secondary here; the primary case is distributional and structural. Say so out loud rather than inflating the score.

The opposition's best case

1. "You're destroying the most successful behavioral intervention in economics." (Madrian & Shea 2001; Thaler & Benartzi 2004; the entire auto-enrollment literature; SECURE 2.0 just mandated auto-enrollment for new plans starting 2025.) This is the strongest objection and it is half right. Auto-enrollment works, and a plank that abolished the 401k and left a voluntary IRA would reduce saving among exactly the workers we claim to help. Concession in writing: the plank is defensible only bundled with a mandatory universal payroll auto-IRA. The response: the behavioral machinery — payroll deduction, default-in, default fund, auto-escalation — attaches to the paycheck, not to the 401k statute, and state auto-IRAs demonstrate it transplants. Choi et al. (NBER w32828, 2024) further show the 401k version's net effect (+0.6% of income) is far smaller than its reputation, largely because of job-tied leakage the IRA design eliminates.
2. "Workers lose the employer match — free money averaging 4.7% of pay." (Brian Graff, American Retirement Association, responding to Biggs–Munnell: "absurd to take away the incentives from a system that's actually working" — 401(k) Specialist, 2024.) Response: matches are compensation, not charity; employers can deposit the same dollars into employee IRAs (SEP-style) or wages, now vesting immediately. But incidence is genuinely uncertain and slow — concession: some workers, especially in slack labor markets, will lose part of the match value in transition. Counterpoint from the leakage study: match-heavy balances are cashed out more at separation (Yin et al., 2023), so some of the match's paper value never survives contact with a job change today.
3. "The tax-expenditure numbers you brandish are inflated." (Peter Brady, ICI; Mercatus Center, "Social Security Needs Fixing, Tax Increases on Savings Are Not the Solution," 2024.) Deferral is not exemption; present-value cost is far below the $251B cash-flow figure. Concession: correct. Our fiscal section uses the cash-flow number only as the official baseline and does not claim it as recoverable revenue.
4. "Forcing $10.1T through rollovers will feed the IRA rollover machine's worst behaviors" — higher-fee advised products, conflicted advice, loss of ERISA fiduciary protection and menu curation (see the trade-press "$9 trillion rollover trap" critique, Kiplinger, 2025; this echoes GAO's long-standing rollover-conflict findings). Concession: today's IRA market has weaker investor protection than ERISA plans, and this is the hardest engineering problem in the plank. Response: the transition must ship with (a) a default in-kind rollover to a low-cost default fund, (b) a TSP-style public default option, and (c) extension of fiduciary/fee-disclosure standards to IRA providers. Without those, don't do it.
5. "IRAs leak worse than 401ks — money can be withdrawn (with penalty) any day, and there are no plan loans." Response: the biggest current leak — the job-separation cash-out (41.4%) — exists because of the employment tether and disappears with it. Penalty rules (IRC §72(t)) carry over unchanged. Evidence on net leakage in an all-IRA world is genuinely thin (estimate, weak evidence). State auto-IRA withdrawal rates are worth monitoring as the test.

Talking points

• One line: Your retirement account should belong to you, not to your job.
• Thirty seconds: Nearly half of private-sector workers — 57 million people — get no retirement plan at work, and the ones who do can shelter $24,500 while everyone else is capped at $7,500 in an IRA. That's the same taxpayer, treated three-to-one differently, because of who signs their paycheck. Give every American one account they own for life, raise its limit to the 401k limit, and auto-enroll everyone through payroll — which Oregon and California are already doing.
• Two minutes: Add the villain and the honest cost. The employer channel supports a $13-trillion administration industry — Fidelity alone runs $2 trillion of 401k money — whose fees hit small-firm workers hardest: about 1.1% a year in small plans versus 0.27% in big ones, and a Yale study found that in 16% of plans, fees ate the entire tax benefit. FSAs quietly transfer about $4.5 billion a year of workers' own money to employers through use-it-or-lose-it. And when people change jobs, 41% cash out their 401k and pay the penalty. Meanwhile 58% of the $250-billion-a-year retirement tax break goes to the top-earning fifth. The fix: end the 401k, FSA, and HSA; move the whole tax break into a bigger IRA every worker owns; auto-enroll everyone through payroll, as 15 states already do. The honest costs: high earners lose shelter above $24,500 — that's the point — the employer match has to convert to wages and that transition is uncertain, and rolling $10 trillion into IRAs safely requires a public low-cost default option and fiduciary rules for IRA providers. Every behavioral trick that made the 401k work — auto-enrollment, defaults, escalation — comes along to the IRA. What doesn't come along is the part where your retirement dies a little every time you change jobs.

Open questions

• Match incidence. How much of the 4.7%-of-pay average employer match actually converts to wages or IRA deposits, and how fast? This is the number an economist should be asked to pin down first; if pass-through is weak, median workers could be net losers for years.
• Default rate adequacy. State auto-IRAs default around 5% with no employer money; 401k combined contributions average 12.1% (Vanguard, 2025). Does the plank need mandatory auto-escalation to a double-digit default?
• Leakage in an all-IRA world. IRAs permit penalized withdrawal anytime; job-separation cash-outs disappear. Net effect unknown — OregonSaves/CalSavers withdrawal data is the natural experiment to watch.
• Nondiscrimination rules. 401k top-heavy/ADP testing forces some employers to fund lower-paid workers' accounts. Repeal removes that lever; the Saver's Match (2027) and auto-IRA mandate must fill the gap. Verify the magnitude.
• Dependent-care FSA has no IRA analog; needs a direct-subsidy replacement (out of scope for this doc).
• Roth treatment, ERISA creditor protection for IRAs, and the fate of defined-benefit plans (untouched by this plank) need explicit statutory answers.
• What evidence would change our mind: credible evidence that auto-IRA participation/contribution levels cannot reach within ~2 points of employer-plan auto-enrollment outcomes at scale, or that match pass-through to wages is near zero over a 5-year horizon.

Sources

1. IRS, "401(k) limit increases to $24,500 for 2026; IRA limit increases to $7,500," November 2025 — primary. (2026 limits; 2025 comparators $23,500/$7,000 stated therein. 2026 HSA $4,400/$8,750 per Rev. Proc. 2025-19 and FSA $3,400/carryover $680 per the IRS 2026 COLA announcements, cross-checked via ALPA benefits summary, January 2026 — secondary.)
2. BLS, Employee Benefits in the United States, March 2025 (USDL-25-1464), September 2025 — primary. (72% access / 53% participation; 59% small-establishment access.)
3. AARP Public Policy Institute, "Payroll Deduction Savings Programs Improve Retirement Security," 2022 (updated) — secondary. (57M without access; 81% earning ≤$50k; 15x payroll-deduction effect.)
4. Ayres & Curtis, "Beyond Diversification: The Pervasive Problem of Excessive Fees and 'Dominated Funds' in 401(k) Plans," Yale Law Journal 124:1346, 2015 — primary (peer-reviewed; 2010 plan data).
5. BrightScope/ICI, "The Defined Contribution Plan Profile: A Close Look at 401(k) Plans," 2024 edition — primary-adjacent industry data. (0.81% average total plan cost, 2021; 1.13% vs 0.27% by size, 2014.)
6. JCT, "Estimates of Federal Tax Expenditures for Fiscal Years 2024–2028," JCX-48-24, December 2024 — primary. ($251.4B DC-plan expenditure, 2024.)
7. Bipartisan Policy Center, "Who Benefits from Retirement Tax Breaks?" (2019 data) — secondary; CBO, "The Distribution of Major Tax Expenditures in 2019," 2021 — primary.
8. Money/EBRI, FSA forfeiture analyses, 2023–2025 — secondary. ($4.5B in 2023; $5.1B in 2022; $441 average.)
9. Yin, Bhattacharya et al., "Cashing Out Retirement Savings at Job Separation," Marketing Science, 2023 — primary. (41.4% cash-out.)
10. GAO-25-107480, "Health Savings Accounts: Information on Features and Use, and Characteristics of Account Holders," 2025 — primary; CBPP on GAO HSA data, 2008 — secondary.
11. KFF, 2025 Employer Health Benefits Survey — primary. (HDHP/SO 33% of plan enrollment, 2025.)
12. Devenir, 2025 Year-End HSA Research Report, early 2026 — secondary (industry survey). ($174B / 41.7M accounts.)
13. ICI, Quarterly Retirement Market Data, Q4 2025 (released March 2026) — primary-adjacent. ($10.1T in 401k; $19.2T in IRAs; $49.1T total.)
14. PLANSPONSOR, 2025 Recordkeeping Survey — secondary. (~$13T DC / ~140M participants; Fidelity $2.04T AUA / 31.7M participants via industry rankings.)
15. OpenSecrets, Investment Company Institute lobbying profile; American Benefits Council profile; 2025 industry totals — secondary (compiled from primary Senate lobbying disclosures).
16. Madrian & Shea, "The Power of Suggestion: Inertia in 401(k) Participation and Savings Behavior," QJE 116(4), 2001 — primary.
17. Thaler & Benartzi, "Save More Tomorrow," Journal of Political Economy 112(1), 2004 — primary.
18. Beshears, Choi, Laibson et al., "Smaller than We Thought? The Effect of Automatic Savings Policies," NBER WP 32828, August 2024 — primary (working paper).
19. Vanguard, "How America Saves 2025" (25th edition, 2024 plan-year data) — primary-adjacent industry data. (94% vs 64% participation; 4.7% average match; 12.1% combined rate.)
20. Biggs & Munnell, "The Case for Using Subsidies for Retirement Plans to Fix Social Security," Center for Retirement Research at Boston College, January 2024 — primary (policy paper; $121B revenue estimate).
21. ASPPA, state auto-IRA program trackers, October 2025–March 2026 — secondary. (CalSavers/OregonSaves figures; 17 states enacted / 15 live / $2.75B.)
22. ICI, "Rollovers Fuel Multi-Trillion IRA Market," 2025 and InvestmentNews on Cerulli rollover projections, 2026 — secondary. ($670B in 2022; ~$941B projected 2026.)
23. Pew Charitable Trusts, "Federal Saver's Match, Coming in 2027," September 2025 — secondary.
24. IRS/Treasury proposed regulations on SECURE 2.0 mandatory automatic enrollment, January 2025 — secondary summary of primary rulemaking.
25. Mercatus Center, "Social Security Needs Fixing, Tax Increases on Savings Are Not the Solution," 2024 — secondary (opposition source).
26. 401(k) Specialist, "Economists Reject Plan to End 401(k) Tax Benefits," 2024 — secondary (Graff/ARA quote).
27. Brady, Peter, "The Tax Benefits and Revenue Costs of Tax Deferral," Investment Company Institute, 2012 — unverified in this research pass (cited from memory; verify before "reviewed" status).
28. IRS Notice 2025-67 (2026 §415(c) limit of $72,000) — unverified direct text; figure taken from secondary coverage of the November 2025 IRS COLA announcement.