The ACA Cliff Costs Early Retirees an Average of $213,290. Our Monte Carlo Shows It's Entirely Avoidable.

QuantCalc Research · Published April 11, 2026 · Engine aca-cliff-mc-2.0.0 · 80,000 simulated paths

The repayment cap is gone. One dollar of modified adjusted gross income over 400% of the federal poverty level in 2026 can claw back the entire annual premium tax credit. We ran 80,000 Monte Carlo paths to quantify the damage — and measure how much of it planning avoids.

Why we ran this

The enhanced premium tax credits enacted in 2021 (ARPA) and extended through 2025 (IRA) had temporarily removed the ACA's 400% FPL subsidy cliff. They expired at the end of 2025 — so for the 2026 coverage year the ACA reverts to its original structure: a hard 400% FPL eligibility cliff, with no cap on repaying advance credits when your income lands above it. (The 2025 reconciliation law separately tightened repayment-cap guardrails below 400% FPL.) The upshot for early retirees: cross 400% FPL and the full annual subsidy can be clawed back at tax time. (Congress has debated restoring the enhanced credits — a House-passed extension remains stalled in the Senate as of mid-2026. If an extension is enacted, the cliff mechanics modeled here would change; this page reflects the law in force for the 2026 coverage year.)

Most planning content on the ACA cliff is written as though it's a new problem to research. It isn't. What's new is the scale of the penalty. An early retiree who triggers the cliff by $1 in MAGI can owe back the full year's subsidy — often around $8,000 for a single retiree or $20,000 for a couple buying the Second Lowest Cost Silver Plan (SLCSP). Stack that over a 10-year pre-Medicare bridge and the cumulative cost of a few bad tax years can erase a sizeable portion of the portfolio.

We wanted numbers, not narrative. So we built a simulation engine, picked four representative early-retiree profiles, modeled two withdrawal behaviors per profile, and ran 10,000 Monte Carlo paths on each — 80,000 paths in total. What follows is the output of aca-cliff-mc-2.0.0.

What we modeled

Each profile represents a 55-year-old household beginning a 10-year bridge between early retirement and Medicare eligibility at 65. All four profiles use the same portfolio allocation, inflation assumption, and account mix. The only things that vary across profiles are household size, starting portfolio value, and annual bridge spending.

For each profile we simulated two behaviors:

• Naive behavior — Withdraw from pre-tax (Traditional IRA/401(k)) first until empty, then Roth, then taxable. This is the default order suggested by most legacy retirement planning rules of thumb.
• Tax-optimized behavior — Prioritize taxable-account withdrawals (capital gains at ~50% basis) and Roth distributions during the pre-Medicare bridge, keeping MAGI below 400% FPL and preserving Traditional IRA assets for post-Medicare spending when ACA exposure ends.

Profile-by-profile results

The table below shows the full output for each profile. pct_cliff_years_1_3 is the share of Monte Carlo paths that cross the 400% FPL threshold in the first three bridge years. mean_total_repayment is the mean cumulative subsidy clawback over all 10 bridge years. The "Value of Planning" column is the mean-total-repayment difference between the two behaviors — the dollars planning actually saves.

Reading the numbers

A few observations from the output block that readers should internalize:

• Cliff exposure in years 1–3 is deterministic, not probabilistic. Every naive path crosses the threshold early because naive withdrawals pull fully taxable distributions straight into MAGI. This is not a sequence-of-returns artifact. It's a behavior artifact.
• Mean repayment is capped near the SLCSP ceiling. For couples, naive mean repayment clusters around $225k over 10 years — approximately $22.5k per year, consistent with a national SLCSP of ~$28.8k minus expected contribution. Singles cluster around $90k — approximately $9.0k per year.
• Optimized "ever crossed" rates are nonzero. Even with good behavior, 31.7–61.7% of paths eventually touch the cliff in at least one year — usually in later bridge years as taxable-account basis is exhausted and forced distributions begin. The mean repayment on those paths is still small because the crossing is typically late and partial.
• Value of planning scales with household size. Singles see roughly $77k–$86k in average savings; couples see $198k–$213k. The couple multiplier is roughly 2.5×, slightly higher than a straight household-size doubling because the couple SLCSP is not exactly 2× the single SLCSP.

Methodology

Full reproducibility details. The engine is versioned and the inputs are public. The engine (research/aca_cliff/compute.py) and its fixed seed are committed to the repo, so anyone can reproduce these figures exactly. The full simulation output is published as open data under a CC0 public-domain dedication: results.json.

We model returns parametrically — long-run real means, volatilities, and a full correlation matrix across all five asset classes — rather than forward-looking Capital Market Expectations, because sampling correlated Monte Carlo paths requires a complete covariance structure. Public forward-looking CME publications (BlackRock, J.P. Morgan, Vanguard, GMO, Schwab, Invesco, Morningstar) report expected returns but rarely publish full covariance matrices. See our full methodology page for the broader treatment of CME sources and why we cross-check them.

Withdrawal-order logic is simplified on purpose. "Naive" pulls from pre-tax until empty; "optimized" pulls from taxable and Roth first and saves pre-tax for post-65. Real-world optimization is richer — partial Roth conversions, IRMAA planning, deferred capital gains, state tax — but adding those would mostly widen the gap between naive and optimized outcomes, not close it. We kept the optimized behavior conservative so the savings numbers are a floor, not a ceiling.

What this means for planners and retirees

If you are within five years of early retirement and you have not explicitly modeled MAGI against the 400% FPL threshold for each bridge year, this simulation is the number you should be running. The default advice — "just withdraw from your 401(k) first" — quietly costs the typical couple the equivalent of a paid-off starter home.

The planning adjustment is not complicated:

• Map your expected MAGI in year 1 of retirement. Include every dollar: Traditional withdrawals, Roth conversions, capital gains, interest, dividends, business income.
• Know your 400% FPL number for your household size in the year you retire. For 2026 that's $62,600 (single) or $84,600 (couple).
• Build a bridge plan that sources as much spending as possible from taxable-account capital gains (where the cost basis dampens MAGI impact) and tax-free Roth distributions.
• Delay Roth conversions until after Medicare eligibility, or size them carefully to stay under the 400% cliff — and remember that the IRMAA threshold at 65 becomes the next cliff to manage.
• Use a real Monte Carlo — not a deterministic spreadsheet — to test the plan across return sequences, because sequence-of-returns risk interacts with the cliff in non-obvious ways.

QuantCalc builds two free tools for this exact workflow. The ACA Cliff Calculator lets you enter MAGI components, household size, and income sources to see whether you cross the 2026 threshold and what it would cost. The Stress Test tool runs 50–10,000 Monte Carlo paths on your actual portfolio and withdrawal plan. Both are free and neither stores your inputs — the ACA Cliff Calculator runs in your browser, while the Stress Test sends inputs over HTTPS to run the simulation and doesn't retain them.

Frequently asked questions