[ER-01] Safe Withdrawal Rates

Five credible studies put the safe rate between 2.31% and 5.50% — a 2.4x spread on the same portfolio. The number you adopt is mostly a statement about which question you asked.

Pillar: Early Retirement Mechanics · Applies to: Anyone converting a portfolio into a spending number, at any age — and anyone who has been given a single figure without its assumptions Last verified: August 2026 · Refresh cadence: Annual — the forward-looking estimates are re-published yearly; the historical studies are fixed Related: ER-06 Sequence of Returns Risk · ER-02 Roth Conversion Ladder · ER-05 Accessing Money Early · TX-07 Guardrails · TR-01 Five-Year Runway · SS-01 Claiming Age Math · [IN-05] Annuities

Not advice. Every rate below is a published research finding with stated assumptions, reproduced here with those assumptions attached. None of them is a prediction, and none was derived for your horizon, your country weighting, or your tolerance for running out. Read §2 before adopting any single number.

---

TL;DR

• On a $1,500,000 portfolio, the published “safe” rates imply first-year spending between $34,650 and $82,500 — a 2.4x spread. The two most-cited current figures sit $47,850 a year apart.
• The 4% rule is a finding about the United States, 1926 onward, over exactly 30 years. Widen the sample to 38 developed markets and the rate falls to 2.31% ✅; the rule was “safe” in only 4 of 17 countries over 109 years ✅.
• Horizon alone does most of the damage, before any risk is modelled. At a guaranteed 3% real return, the sustainable rate falls from 5.10% over 30 years to 3.89% over 50 — a 24% cut with no uncertainty in the model at all.
• “Success” is a binary that hides everything. A plan running dry in year 28 delivered $1,680,000 of real spending; one failing in year 12 delivered $720,000. Both are scored identically.
• Every study assumes a retiree who never adjusts. That assumption is what makes the published rates conservative, and adjusting is worth more than any allocation choice (ER-06 §5).

---

1. The rates, in dollars

Rate	First-year spending on $1.5M	What produced it
2.31%	$34,650	Anarkulova, Cederburg, O’Doherty & Sias — 38 developed markets, 5% ruin risk ✅
3.90%	$58,500	Morningstar 2026 — forward-looking, 30 years, 90% success, 30–50% equity ✅
4.15%	$62,250	Bengen 1994 — worst US 30-year period in the historical record ✅
4.70%	$70,500	Bengen 2025 — “Universal SAFEMAX,” broader asset classes ✅
5.50%	$82,500	Bengen 2025 — under current valuations and inflation ✅

None of these numbers is wrong and nobody in the table is careless. They answer different questions, and the question is the input a retiree is never shown. The gap between the top and bottom rows — $47,850 a year — is not a rounding difference. It is the difference between a comfortable retirement and a frugal one, and it is decided entirely by assumptions made before the arithmetic starts.

2. What each rate is actually claiming

Sample	Horizon	Failure bar	Rate
US only, historical	30 years	worst case observed	4.15%
US only, historical, more asset classes	30 years	worst case observed	4.70%
US, forward-looking returns	30 years	10% chance of failure	3.90%
38 developed markets	to mortality	5% chance of ruin	2.31%

Three levers, and each moves the answer more than most people expect.

The sample. Bengen and Trinity used US data from 1926. The United States had the best-performing large equity market of the twentieth century, and using its record to calibrate a rule is survivorship bias with a spreadsheet attached. Pfau tested the 4% rule across 17 developed countries over 109 years and found it “safe” in only 4 of them ✅ — it failed in the majority of cases in Italy, Spain, Belgium, Germany, and France ✅. The 2022 working paper by Anarkulova, Cederburg, O’Doherty and Sias extends this to 38 countries with block-bootstrap resampling and lands at 2.26%, published in the Journal of Pension Economics & Finance in July 2025 at 2.31% ✅.

The horizon. 30 years is a 65-year-old’s number. §3.

The failure bar. “Worst case observed” is not the same as “5% chance,” and neither is “90% success.” Morningstar’s own framing is the one to keep: a 90% success rate is a 10% failure rate, and the failures cluster in exactly the bad-sequence paths that ER-06 is about.

The honest summary: the 4% rule is a fact about one country’s best century, presented as a fact about retirement. That does not make it useless — the US is where this audience actually holds assets and earns returns — but it makes a single decimal place false precision.

3. Horizon does most of the damage, before any risk is modelled

Strip out uncertainty entirely. Assume a guaranteed real return and ask what a portfolio can pay:

Guaranteed real return	30 years	40 years	50 years
2%	4.46%	3.66%	3.18%
3%	5.10%	4.33%	3.89%
4%	5.78%	5.05%	4.66%

At a guaranteed 3% real return, the sustainable rate falls from 5.10% to 3.89% going from 30 years to 50 — a 24% cut in spending with no risk in the model whatsoever.

This is the cleanest way to see why FIRE cannot borrow a traditional retiree’s number. The reduction is arithmetic, not caution. Sequence risk (ER-06) then stacks on top of it, and inflation compounding over fifty years stacks on top of that. A 45-year-old using 4% because a 65-year-old’s study said so has made two independent errors that happen to point the same way.

The corollary is more useful than the warning: anything that shortens the portfolio-dependent horizon is worth more than anything that optimises the rate. Delaying Social Security converts the last third of the horizon into indexed income (SS-01); a pension or an annuity floor does the same at a price ([IN-05]). The strongest move available is usually to make the horizon shorter, not the rate higher.

4. The failure bar hides what a retiree cares about

Two plans, both spending $60,000 a year in real terms. Every study on this page scores both as failures:

	Real spending delivered
Ran dry in year 28	$1,680,000
Ran dry in year 12	$720,000

A 95% success rate tells you nothing about which of these you got, and the difference between them is most of what matters. Worse, both figures assume a household that kept spending at full rate into the wall without adjusting — which no real household does, and which is the single largest reason published rates understate what is actually survivable.

This is why “what is the safe rate” is the wrong question and “what is my plan when it goes badly” is the right one. A retiree with a written rule for cutting spending after a bad year is running a fundamentally different plan from the one these studies model, and ER-06 §5 prices that difference: on the same returns, a persistent 10% cut after down years turned a near-failure into a large surplus — while real spending fell 42% at the trough. The flexibility is worth a great deal and it is not free.

5. The variants, and what each is for

Approach	What it does	Best for
Fixed real (the 4% rule)	Withdraw X% in year one, inflate it thereafter	Modelling and comparison. Almost nobody should actually run it
Fixed percentage	Withdraw X% of the current balance each year	Never runs out by construction; spending swings with the market
Guardrails (Guyton-Klinger and successors)	Fixed real, with rules that cut after bad years and raise after good ones	Most retirees — the compromise that keeps spending stable but bounded (TX-07)
RMD / VPW method	Divide the balance by remaining life expectancy	Simple, self-correcting, automatically front-loads spending
Floor and upside	Cover essentials with indexed income, take risk only with the surplus	Anyone whose fixed costs are a large share of spending ([IN-05], SS-01)

The choice among these matters more than the starting rate, and it is the part almost never discussed alongside the headline number. TX-07 owns the specific guardrail rules; what belongs here is the ranking: a household able to vary spending should start higher than one that cannot, and Morningstar’s own model agrees — the same research that yields 3.90% for a rigid plan yields roughly 5.7% ◻️ where spending adjusts.

6. Two inputs people argue about, ranked by how much they matter

Allocation matters, but less than the arguments about it suggest. Trinity found 4% sustainable at both 50/50 and 75/25 stock/bond ✅ — a wide plateau rather than a peak, which is the useful finding. Below roughly 40% equity the portfolio stops outrunning inflation over long horizons; above roughly 75% the historical gain in sustainable rate flattens while sequence exposure keeps rising (ER-06). Anywhere in that band is a defensible answer, and time spent optimising inside it is time not spent on the horizon and the spending rule, both of which move the number more.

Note what Morningstar’s 3.90% assumes: 30–50% equity ✅, which is lower than most retirees hold. Its rate is not directly comparable to Bengen’s for that reason alone — a reminder that “the safe rate” is never one number even within a single study.

Valuation is the live disagreement. Bengen’s current 5%–5.5% range is explicitly conditioned on today’s CAPE and inflation ✅, and Pfau and Kitces’s glidepath work finds the optimal path becomes more conservative in low-expected-return environments ✅. These point in opposite directions from the same starting observation, which is a fair summary of the state of the field: valuation clearly matters, and nobody agrees on the sign of the adjustment at any given moment.

The practical resolution is not to pick a side. Start nearer the low end of your horizon’s range, write the adjustment rule, and let the rule do the work that a valuation forecast would otherwise have to do. That is the same conclusion §4 reached from the failure-bar direction, arrived at independently.

7. What to actually do with this

1. Pick the rate that matches your horizon, not your hopes. Thirty years is 65+; a 45-year-old should be looking at the 50-year column in §3.
2. Compute the number, then subtract non-portfolio income. The rate applies only to the portfolio-funded part of spending — Social Security, a pension, or rental income shrink the problem before any rate is chosen (SS-01).
3. Write the adjustment rule down now, while markets are calm (TX-07, ER-06).
4. Recompute annually against the current balance, not the original. A 4% plan on a portfolio that fell 30% is a 5.7% plan and nobody sent a notice.
5. Treat any single decimal place as false precision. The honest output is a range with a written rule for moving inside it.

Three Scenarios

A. Retiring at 65 with $1.5M and Social Security starting at 70. Only the bridge years are fully portfolio-dependent; from 70 a large share of spending is indexed and guaranteed. His effective horizon for the portfolio-funded part is much shorter than 30 years, which justifies a higher rate on that slice than any headline number implies — and the bridge itself is where sequence risk concentrates (ER-06, SS-05).

B. Retiring at 42 with $2.5M and no Social Security for 25 years. The 50-year column governs, and even at a guaranteed 3% real return that is 3.89%. Adding sequence risk and 38-country evidence pushes the defensible range lower still. Her strongest lever is not the rate: it is that she has decades of optionality to earn again, and a written rule to cut spending, both of which the studies assume away (ER-06).

C. The couple handed “4%” by an advisor and nothing else. The number may well be reasonable. What is missing is the horizon it assumes, the country it was measured in, the failure probability it accepts, and what they are supposed to do in a bad year — and the fourth is the one that decides the outcome. Asking those four questions is the whole content of this page.

💡 Pro-Tips

1. Ask any quoted rate four questions: which sample, which horizon, which failure bar, and what happens in a bad year.
2. Apply the rate only to portfolio-funded spending. Netting non-portfolio income first often changes the answer more than the rate choice does.
3. Shorten the horizon rather than raise the rate — delaying Social Security is the cheapest way to do it (SS-01).
4. Start higher if you can genuinely flex spending, lower if your budget is mostly fixed. That is the single most informative fact about your own situation.
5. Recompute against the current balance each year. Fixed-real plans silently become high-rate plans after a decline.
6. Do not run a fixed-real rule in practice. It is a modelling device; the variants in §5 are what people should actually live on.
7. Treat 2.31% and 5.50% as the honest ends of the range, not as one credible number and one outlier. Both are peer-reviewed or published work with stated assumptions.
8. Front-load discretionary spending deliberately if you front-load it at all — the RMD and VPW methods do this by construction, and spending capacity in the first decade is worth more than in the fourth.

⚠️ Common Pitfalls

1. Using a 30-year rate for a 50-year retirement. Arithmetic alone takes 5.10% to 3.89% at a guaranteed return.
2. Treating the 4% rule as a law of nature rather than a US-only, 30-year, worst-historical-case finding ✅.
3. Ignoring the international evidence because it is inconvenient — it was “safe” in 4 of 17 countries ✅.
4. Reading “95% success” as “95% fine.” It scores a year-28 failure and a year-12 failure identically (§4).
5. Assuming you will behave like the model — spending unchanged into a 40% decline is what the studies simulate and what nobody does.
6. Applying the rate to total spending including the part Social Security or a pension already covers.
7. Holding the first-year dollar figure fixed after a large decline, which quietly raises the real withdrawal rate.
8. Choosing a rate without choosing a rule. The variant in §5 matters more than the decimal.
9. Believing a higher equity allocation buys a higher safe rate indefinitely — past roughly 50–75% equity the historical gains flatten and sequence exposure rises (ER-06).
10. Confusing a safe withdrawal rate with a sustainable one. “Safe” means it survived a specified bad case; it says nothing about what is likely.

✅ Actionable Checklist

Setting the number

• Write down your horizon in years, from your actual retirement age
• Subtract non-portfolio income from target spending; the rate applies to the remainder (SS-01)
• Pick a rate from §1 whose assumptions you can state out loud
• Sanity-check it against the guaranteed-return table in §3 for your horizon

Choosing the rule, not just the rate

• Pick a variant from §5 and write it down (TX-07)
• Define the trigger, the size of the cut, and how long it persists
• Decide what share of spending is genuinely discretionary — that sets how much flexibility you actually own

Every year

• Recompute the withdrawal rate against the current portfolio value
• Apply the adjustment rule as written, not as it feels
• Re-check the horizon; it shortens by a year, and that is a real loosening
• Confirm the plan still clears the ACA or IRMAA threshold it was built around (HC-01, HC-06)

Sources & further reading (verified August 2026)

• Bengen, W., “Determining Withdrawal Rates Using Historical Data” (Journal of Financial Planning, October 1994) — the original 4.15% SAFEMAX, derived as the worst outcome in the US historical record over rolling 30-year periods.
• Cooley, Hubbard & Walz (Trinity University), “Retirement Savings: Choosing a Withdrawal Rate That Is Sustainable” (1998) — the “Trinity study”: 4% inflation-adjusted withdrawals succeeded in 95–100% of historical 30-year periods at 50/50 and 75/25 stock/bond allocations.
• Bengen, W., A Richer Retirement (Wiley, August 2025) — the revision to a 4.7% “Universal SAFEMAX” with broader asset classes, and 5%–5.5% under current valuation and inflation conditions.
• Pfau, W., “An International Perspective on Safe Withdrawal Rates from Retirement Savings: The Demise of the 4 Percent Rule?” (Journal of Financial Planning, December 2010) — 17 developed countries over 109 years; a 4% real withdrawal was “safe” in only 4 of them.
• Anarkulova, A., Cederburg, S., O’Doherty, M. & Sias, R., “The Safe Withdrawal Rate: Evidence from a Broad Sample of Developed Markets” — 38 developed countries with block-bootstrap resampling; 2.26% in the 2022 working paper, 2.31% in the version published in the Journal of Pension Economics & Finance, July 2025, for a 65-year-old couple accepting a 5% chance of ruin.
• Morningstar, “The State of Retirement Income” (2026 edition) — 3.9% for a 30-year horizon at 30–50% equity and 90% success, with a substantially higher rate available where spending adjusts.
• tools/er01_worked_examples.py — the dollar translations, the horizon table, and the failure-bar comparison. The published rates are inputs, verified above; the arithmetic is computed and checked against this article by CI.
• ER-06 owns sequence risk, which is why these rates are below the average return · TX-07 owns guardrail rules · [IN-05] owns the annuity floor · SS-01 owns the horizon-shortening decision.