[TR-03] One More Year, or Quitting Too Soon

The marginal working year gets more valuable every year you stay — so no test of the margin will ever tell you to stop

Pillar: Approaching Retirement & Transitions · Applies to: Anyone at or near their number who keeps finding a reason to stay another year, and anyone about to leave on a number they have not stress-tested Last verified: August 2026 · Refresh cadence: Evergreen (the arithmetic); figures via FN-02 Related: TR-01 Five-Year Runway · TR-04 Practice Retirement · ER-01 Safe Withdrawal Rates · ER-06 Sequence of Returns Risk · ER-07 Coast and Barista FIRE · SS-04 Earnings Test and Your PIA · HC-01 ACA Bridge · TX-02 Roth Conversion Strategy

Not advice. This page prices a decision; it does not make one. The arithmetic is exact and the assumptions under it — a real return, a planning horizon, a spending number — are yours and are not verifiable in advance. Where this page cites behavioural or empirical research it marks the strength of that evidence, and one of the two findings it leans on is actively contested. Treat the marks as part of the content.

---

TL;DR

• The marginal working year does not diminish. It accelerates. For the reference household in §1, the first extra year adds $7,062 of sustainable spending and the fifth adds $9,674 — the marginal year is worth 37% more after five years of waiting.
• That is the whole diagnosis. “One more year syndrome” is not a failure to notice diminishing returns; there are none to notice. It is the correct response to a stopping criterion that can never be satisfied. You cannot exit on a test of the margin — only on a threshold set in advance.
• The cost side accelerates faster. A working year spends a finite, non-renewable stock of cheap-conversion and subsidised-coverage years, from the cheap end first. Retiring at 58 makes $323,000 of traditional balance movable under the ACA cliff; working to 63 makes none of it movable.
• The two errors are not symmetric, and the asymmetry is computable. Two years of “quitting too soon” opens a $266,500 funding gap that two years of returning to work closes — in a good market and a bad one. The conversion years and the healthy years do not come back on any schedule.
• Past 35 years of covered earnings, “one more year” is not a Social Security argument at all — it adds $0 to your benefit at the same pay ✅ (SS-04 §3).

---

1. The marginal year accelerates

Take a household that is genuinely close: both 58, portfolio $1,600,000, spending $80,000 a year, still saving $50,000, planning to 95.

Sustainable spending here is the ER-01 §3 benchmark — amortise the portfolio over the remaining years at a guaranteed 3% real return. That is deliberately not a withdrawal-rate recommendation; ER-01 found credible published rates spanning 2.31% to 5.50% ✅ and this page takes no position among them. A risk-free benchmark is used because the question here is comparative — this year against next year — and removing return uncertainty isolates the thing being measured. Accumulation grows at a stated 5% real.

Retire at	Portfolio	Horizon	Sustainable	Funded	That year added
58	$1,600,000	37 yrs	$72,179	90.2%	—
59	$1,730,000	36 yrs	$79,241	99.1%	$7,062
60	$1,866,500	35 yrs	$86,866	108.6%	$7,625
61	$2,009,825	34 yrs	$95,109	118.9%	$8,243
62	$2,160,316	33 yrs	$104,032	130.0%	$8,924
63	$2,318,332	32 yrs	$113,706	142.1%	$9,674

Read the last column downward. Every additional year of work is worth more than the one before it: $7,062 rising to $9,674, a 37% increase in the value of the marginal year over five years of waiting.

Three effects push the same way and none of them exhausts:

• The portfolio compounds, so a fixed savings rate lands on a larger base each year.
• The horizon shortens, so the same portfolio supports a higher draw — the ER-01 §3 effect run forwards.
• A year not spent is a year not withdrawn, worth as much as a year saved and invisible on most spreadsheets.

This is what the popular framing gets backwards. It treats the syndrome as an inability to see that the extra year buys little. For a household still earning and still saving, the extra year buys a lot, and more each time. The person staying another year is not miscalculating. They are calculating correctly and applying a test that has no stopping point.

2. Why that makes it a criterion, not a decision

A stopping rule is only a rule if some future state falsifies it. Compare:

Stopping rule	Terminates?	Why
“When another year stops being worth much”	No	§1 — the quantity it watches is rising
“When the markets settle down”	No	No defined settled state; every level has a case for waiting
“When I have a comfortable cushion”	No	“Comfortable” is defined by the current balance and moves with it
“When the kids are sorted”	No	No terminal condition; the definition of sorted extends
“At 110% funded on the §1 benchmark”	Yes	Crossed at 60 in the table above
“On 15 March, when the equity tranche vests”	Yes	A date (TR-01 §4)
“When the bridge account reaches $440,000”	Yes	A balance (TR-01 §7)

The diagnostic is one question: what observable event ends this? If you cannot name the event, you have not made a decision — you have adopted a policy of working, and policies do not expire.

That reframing matters because it is checkable by someone else. “Am I rationalising?” is unanswerable; “does my stated rule have a terminal condition?” is answerable in a sentence — and answerable by a spouse, which is where the conversation usually needs to happen ([LS-04]).

Both errors on this page are one test. Working too long is a rule with no terminal condition. Quitting too soon is a rule whose terminal condition was computed wrong (§5). Neither is a personality trait.

3. The other ledger: what the marginal year spends

§1 counted only what the year adds. The year also consumes something, and that stock is finite.

Two windows open at retirement and close on fixed dates regardless of what your portfolio does:

• Unconstrained conversion years run from the retirement year through the year you turn 62. From 63 on, every dollar of MAGI also sets a Medicare premium two years out ✅ (TR-01 §6, HC-06).
• ACA bridge years — the years your MAGI is genuinely chooseable — run until Medicare at 65 (HC-01).

For the §1 household: spending funded from taxable basis and Roth basis throws off perhaps $20,000 of unavoidable MAGI, leaving $64,600 of headroom under the $84,600 ✅ 400% FPL cliff for a household of two (FN-02, HC-01).

Retire at	Unconstrained conversion yrs	ACA bridge yrs	Traditional balance movable
58	5	7	$323,000
59	4	6	$258,400
60	3	5	$193,800
61	2	4	$129,200
62	1	3	$64,600
63	0	2	nothing

The window is consumed from the cheap end first, and the losses are not proportional. The first year of waiting costs one year out of five — 20% of the window. The fifth costs the only one remaining — 100% of what is left. So the cost of the marginal year rises too, and faster than its value rises.

Two honest qualifications. Converting to the cliff and claiming a large subsidy are substantially rival strategies — the conversion is itself MAGI, so filling the window carries a premium cost that HC-01 §4 prices; you are generally a subsidy household or a conversion household, not both (TR-01 §3, ER-02). And the window is worth using, not maximising: a conversion cannot be undone, so HC-01 says undershoot the line.

What survives both qualifications is the shape. Whatever a bridge year is worth to you, you have a fixed number of them, the number is set by your retirement date, and the last few working years destroy them fastest.

The Social Security ledger runs the same way and is usually assumed to run the opposite way. A marginal working year adds $91 a month to the benefit after 20 years of covered earnings, $43 after 30, and exactly $0 after 35 at the same pay ✅ (SS-04 §3). The household most likely to be weighing “one more year” — a long-tenured high earner in their late fifties — is precisely the household for whom the Social Security argument is worth nothing.

4. The two errors are not symmetric

The usual framing treats this as a balanced risk: work too long and waste time, quit too soon and run out of money. The arithmetic does not support the symmetry.

Take the §1 household and let them get it wrong in the “quit too soon” direction: they retire at 58 and decide two years later that they left early. The same market path is applied to both sides, because a comparison that gives one path a better market is not a comparison.

Ordinary market — 5% real, twice:

	At 60
Worked straight through	$1,866,500 — 108.6% funded
Retired at 58	$1,600,000 — 93.1% funded
Gap the error opened	$266,500

Note the second row before the third. Two years of retirement in an ordinary market left them better funded than the day they quit — 93.1% against 90.2% — because the horizon shortened faster than the portfolio drew down. The error is a gap against the alternative, not a decline.

Returning to work at 60 closes it:

Back at work	Retire at	Portfolio	Funded
1 year	61	$1,730,000	102.3%
2 years	62	$1,866,500	112.4% — restored

Bad early sequence — −15% real, then 5%:

	At 60
Worked straight through	$1,530,500 — 89.0% funded
Retired at 58	$1,264,000 — 73.5% funded
Gap the error opened	$266,500

Back at work	Retire at	Funded
1 year	61	81.5%
2 years	62	90.1% — restored

The reversal cost is two years in both worlds, and the gap is identical to the dollar. That is not a coincidence: the gap is forgone saving plus taken withdrawals, compounded — so it moves with the market only weakly over a short error, and a bad sequence damages the household that stayed at work almost as much (ER-06).

The asymmetry is therefore not where people put it. Two years of leaving early cost roughly two years of returning. Two years of leaving late cost two conversion windows that no later decision reopens, and two years of healthy time that nothing reopens at all. One error is priced in years of work. The other is priced in years of life, and only one of those is a renewable resource.

The rule that falls out: make the error you can undo. Not a licence to quit unprepared — §5 is the constraint — but where the two directions are genuinely close, they are not equally recoverable, and the recoverable one is not the default.

5. Quitting too soon is a different failure: the number was wrong

The mirror error is rarely a failure of nerve. It is almost always a funding ratio computed on an input that does not survive contact.

Six ways the number comes out flattering, each owned by a page that prices it:

1. A withdrawal rate borrowed from a 30-year study for a 40-year retirement. Horizon alone cuts the sustainable rate from 5.10% to 3.89% going from 30 years to 50 at a guaranteed 3% real ✅ (ER-01 §3).
2. Healthcare priced at today’s employer contribution. The 2026 cliff is back and unlimited clawback with it (HC-01); the bridge is frequently the largest line in an early-retirement budget.
3. Spending measured in a year that was not typical — no roof, no car, no adult-child event.
4. Sequence risk unmodelled. The §4 bad-sequence column is the mild version (ER-06).
5. Taxes on the drawdown ignored, because the accumulation-phase habit is to think in gross balances (TX-03).
6. The portfolio counted, the access ignored. A funded retirement locked behind 59½ is not funded (ER-05, ER-02).

And one input that runs the other way, which is why it deserves its own mark. Real spending is usually assumed flat for thirty years, and there is evidence it declines: Blanchett (2014) found real retiree expenditure falling roughly 1% a year, troughing near 26% below the starting level around age 84 ◻️. That finding is contested and the mark is the point — it rests on cross-sectional patterns, and Tharp (2026) argues within-household dynamics do not reproduce them ◻️. Do not build a retirement date on it. It is a reason not to over-assume the flat case, not a reason to shave the number.

The test that resolves most of this is not arithmetic. Live on the retirement budget for six months while still earning (TR-01 §3), or take the trial run properly (TR-04). A ratio built on a rehearsed budget is a different object from one built on a projection.

6. A stopping rule that terminates

Four properties. A rule missing any one of them is a criterion, not a decision.

1. It names an observable. A balance, a funding ratio, a date, a vest — something a third party could check.
2. It is written down before the year in which it fires. A threshold set while standing at it is not a threshold.
3. It has a review clause, not an override clause. “Re-run annually against the actual balance” (ER-07 §2) is a review. “Unless things feel uncertain” is an override, and an override with no terminal condition reintroduces §2.
4. It names the second decision too — what happens if the number is missed. “Work to 61 instead,” or “cut spending to $72,000,” or “go part-time” (ER-07, [IN-04]) beats leaving it open, because an open failure branch is what converts a threshold back into a policy.

The threshold should be a funding ratio, not a portfolio balance. A balance goes stale against spending, inflation, and horizon; a ratio re-prices itself every year and is the same number your review clause already computes.

And state the withdrawal assumption, because it moves the answer more than another year of work does. The §1 household is 90.2% funded at the guaranteed-3% benchmark. At ER-01’s most conservative published rate they are nowhere near ready; at its most aggressive they were ready two years ago. The rate you adopt is a bigger lever than the date you choose, and picking the date without stating the rate is how this decision is most often made badly — in both directions.

7. What the evidence actually supports

The temptation on a page like this is to reach for psychology, and most of the quotable psychology does not survive checking.

“One more year syndrome” has no peer-reviewed literature under that name. Searching it returns personal-finance blogs and FIRE-community writing, not journals — including the widely-repeated claim that its roots are “psychological, not financial,” which is asserted rather than evidenced ◻️. That absence is why §2 diagnoses the pattern structurally rather than psychologically. A stopping rule either has a terminal condition or it does not; that is checkable without a theory of mind.

What does clear the bar is the reversibility evidence, and it is the most decision-relevant finding on the page. Maestas (2010), using Health and Retirement Study panel data, found that at least 26% of retirees subsequently return to work, and — the part that matters here — that most such returns were anticipated before retiring rather than driven by financial shocks, poor planning, or low wealth ✅. Among unanticipated returns, the driver was a change in preferences rather than a change in income ✅.

That is the empirical counterpart to §4. The fear underneath most one-more-year decisions is that the exit is a one-way door. It demonstrably is not: returning to work is common, usually planned, and not typically a distress event. The genuine one-way doors are a different list, and TR-01 §6 enumerates them — the Rule of 55 destroyed by a rollover, the deferred-comp election, the age-63 MAGI year, the COBRA fork, Medigap open enrollment. Those close permanently. The retirement date mostly does not.

Two things deliberately not cited. Ego depletion and priming are the standard furniture of retirement-psychology writing and both failed large-scale replication ◻️; and consumer “retirement regret” surveys, which would have supplied a quotable percentage for either side of this argument, are marketing research rather than evidence.

The one input this page refuses to supply is the value of a year of your life. That is not modesty — it is the correct treatment. Healthy remaining years is what the decision turns on, it varies enormously by individual, and published disability-free life expectancy is both stale and a population average. Supply it yourself and the arithmetic finishes: judge that you have 20 good years left, and one more working year is 5% of them, five years a quarter. SS-01 §4 carries the corpus’s longevity note — roughly a 50% chance at least one member of a 65-year-old couple reaches 90 ◻️.

Three Scenarios

A. 58, 90.2% funded, has said “one more year” three years running. He is not confused about the arithmetic — §1 says each year genuinely adds more than the last, and he is right that he is safer every year. His problem is that his rule is “when I feel secure,” which §2 shows cannot fire. The fix is one written sentence: 110% funded on the guaranteed-3% benchmark, reviewed each January. The table says that is age 60 — and that the wait cost two of his five conversion years.

B. 61, 142.1% funded, still going. She crossed every plausible threshold years ago and is working through the expensive end of §3, with one unconstrained conversion year left. The marginal year is worth $9,674 of sustainable spending she does not need, against a window she cannot rebuild. Here the arithmetic has stopped being the argument, and TR-04 — a sabbatical or leave of absence — resolves it better than another spreadsheet.

C. 58, 90.2% funded, about to leave on a 5% withdrawal rate. Opposite problem, same missing piece. His ratio is flattering because it borrowed a rate for a horizon it was not derived for (§5.1) and priced healthcare at his payroll deduction (§5.2). He should not reflexively work three more years; he should rebuild the number and then decide. §4 is on his side — if he leaves and is wrong, two years back at work restores it — but that is a reason to act on a corrected number, not to skip correcting it.

💡 Pro-Tips

1. Write the threshold down before the year it fires, in the form of a funding ratio with a stated withdrawal assumption. A number chosen while standing on it is not a threshold.
2. Test any stated reason for staying against one question: what observable event ends it? If nothing does, it is a policy, not a decision (§2).
3. Count your remaining unconstrained conversion years — retirement year through the year you turn 62 ✅ — and treat that count as a depleting asset, because it is (§3).
4. Past 35 years of covered earnings, drop Social Security from the argument entirely. It adds $0 at the same pay ✅ (SS-04 §3).
5. Prefer the recoverable error. Returning to work is common and usually planned ✅; a spent conversion window and a spent year are not recoverable at all (§4, §7).
6. Name the failure branch in advance — work to 61, cut to $72,000, or go part-time (ER-07). An open failure branch turns a threshold back into a criterion.
7. Rehearse the budget for six months before trusting the ratio (TR-01 §3), or take a real trial run (TR-04).
8. Re-run the ratio annually against the actual balance (ER-07 §2). This is the review clause that makes a written threshold safe to hold.

⚠️ Common Pitfalls

1. Waiting for the marginal year to stop being worth it. It never does — it is worth 37% more after five years (§1).
2. Believing the syndrome is irrationality. The arithmetic supports staying; it is the criterion that is broken, not the sums (§2).
3. Counting what the year adds and not what it spends. The conversion and bridge windows are finite and consumed from the cheap end (§3).
4. Assuming one more year improves the Social Security benefit. After 35 years at the same pay it adds nothing ✅ (SS-04).
5. Working through 62 into 63 without noticing the door. From 63 every conversion is also a Medicare premium decision ✅ (TR-01 §6, HC-06).
6. Treating the retirement date as a one-way door. It mostly is not ✅ (TR-01 §6 has the doors that actually are).
7. Quitting on a withdrawal rate borrowed from a shorter horizon — 5.10% to 3.89% is the cost of that borrowing ✅ (ER-01 §3).
8. Pricing healthcare at your current payroll deduction rather than at the marketplace with the 2026 cliff (HC-01).
9. Building the date on a declining-spending assumption. The finding is contested ◻️ and cuts the wrong way to lean on (§5).
10. Setting the threshold as a portfolio balance rather than a funding ratio, so it goes stale against spending and horizon (§6).
11. Letting “quit too soon” and “work too long” be argued as symmetric risks. They are recoverable on completely different terms (§4).

✅ Actionable Checklist

Before the decision is live — 12+ months out

• Compute the funding ratio at a stated withdrawal assumption, and again at ER-01’s conservative and aggressive published rates
• Audit the ratio against the six inputs in §5 — horizon, healthcare, spending year, sequence, drawdown tax, access age
• Rehearse the budget for six months on the retirement number (TR-01 §3)
• Count remaining unconstrained conversion years and bridge years (§3)

Writing the rule

• State the threshold as a funding ratio, with the assumption it is computed under
• Confirm it names an observable a third party could check (§6)
• Add the review clause: re-run annually against the actual balance
• Name the failure branch — work longer, spend less, or go part-time
• Say it out loud to the person it also affects ([LS-04])

If the answer is “one more year”

• Write down the observable event that ends it, or concede there isn’t one
• Price the conversion and bridge years that year consumes (§3)
• Check the year against the one-way doors, especially age 63 ✅ (TR-01 §6)
• Set a date to re-decide, not a feeling

If the answer is “go now”

• Confirm the ratio survived the §5 audit
• Confirm the bridge exists in accounts you can actually reach (ER-05, TR-01 §7)
• Run the whole TR-01 runway, especially the rollover and COBRA decisions
• Decide in advance what would send you back to work, and at what threshold ✅ (TR-04)

Sources & further reading (verified August 2026)

• tools/tr03_worked_examples.py — the marginal-year ledger, the window ledger, and the reversal cost, all reproducible by running the file and checked against this article by CI. The 5% real growth and the guaranteed 3% benchmark are stated inputs, not findings.
• Maestas, N. (2010), “Back to Work: Expectations and Realizations of Work after Retirement,” Journal of Human Resources 45(3): 718–748 — Health and Retirement Study panel evidence that at least 26% of retirees unretire, that most such returns are anticipated before retirement, and that unanticipated returns reflect changed preferences rather than financial shocks ✅. This is the evidentiary basis for §4’s reversibility claim.
• Blanchett, D. (2014), “Exploring the Retirement Consumption Puzzle,” Journal of Financial Planning 27(5): 34–42 — real retiree spending declining roughly 1% a year, troughing near 26% below the starting level in the mid-eighties ◻️. Contested: Tharp, D. (2026), “The Retirement Spending Smile Revisited: Cross-Sectional Patterns versus Within-Household Dynamics,” and Blanchett, D. (2026), “How Spending Evolves in Retirement: A Smile, a Smirk, or Something Else?”, Financial Planning Review, both distinguish cross-sectional patterns from within-household trajectories ◻️. Cited here only as a reason not to over-assume the flat case (§5).
• No peer-reviewed source is cited for “one more year syndrome” because none was found ◻️ — the term appears in personal-finance writing rather than in journals, which is why §2 diagnoses the pattern by the structure of the stopping rule rather than by a named bias.
• ER-01 §3 owns the horizon arithmetic this page’s benchmark reuses, and §2 owns the published-rate spread · SS-04 §3 owns the marginal year’s Social Security value · TR-01 §6 owns the one-way doors and §7 the bridge sizing · HC-01 owns the subsidy cliff and the conversion-versus-subsidy tension · ER-06 owns sequence risk · TR-04 owns trialling the decision instead of arguing it · ER-07 owns the downshift this page treats as a failure branch.