Why the 4% rule is riskier at 40 than at 65

The 4% rule arrived as a 30-year sentence. William Bengen's original question was about a retiree who needed the money to last about three decades, not five. Repeating 4% at 40 is repeating the rate and changing the homework. This page holds the rate and the starting pot fixed, and moves only the horizon.

Specimen: $1,000,000 already retired, rigid 4% ($40,000/year, band 0, adjust 0), 80/15/5, seed 20260622, 1,000 paths. Horizons to age 90: 40 (50 years), 50 (40), 55 (35), 60 (30), 65 (25). Pairing is by seed, not month-aligned paths across different lengths. Primary contrast: 40 versus 65.

Why the 4% rule is riskier at 40 than at 65

Because there are more years for a bad sequence to land, and more years of withdrawals after it lands. On this freeze, success at 65 is 93.5%. Success at 40 is 78.7%. That is not a rounding difference. It is 14.8% of modelled lives that fail the long sentence and pass the short one, on the same rate and the same starting million.

Intermediate ages sit on a slope, not a cliff: 83.5% at 50, 86.2% at 55, 89.6% at 60. Each extra decade of retirement is another decade the 4% has to survive. The original literature never claimed those decades were free.

Median terminals can look kind at 40 because the survivors have more time to compound: $1,574,241 versus $1,021,875. That is the success-versus-balance trap again. The longer horizon raises both the right tail and the ruin mass. p95 at 40 is $24,476,518. p5 is $0 at every age on this table. Quoting the median of the long life without the success rate is how 4% at 40 looks "fine."

What "safe" would have to mean

This article will not say 4% is safe at 65. 6.5% of 25-year lives still fail on this engine. Safe is a forbidden word without a floor, a history, and a flexible rule that has been named. The claim that survived the gate is narrower: rigid 4% fails more often over a 50-year life than over a 25-year life, on this generator, this pot, this seed.

The August article on withdrawal rates by retirement age used 200 GBM paths and a different household. This freeze is 1,000 regime paths and a flat $1,000,000 start. They are not the same study. If the slopes agree, that is interesting. If they disagree, that is also information. Do not average them. Do not cite one as a replication of the other without saying the generator changed.

Early retirement as a lifestyle is not on trial. The 4% identity is. Horizon risk is the extra years, not a different personality. A 3% rigid rate would have a different slope. A Guyton-Klinger band would hide some failures in spending cuts. Those are other cards. This card froze 4% rigid so the horizon could be the only moving part.

Q2: the shave did not flip the order

Stock mu minus 0.02, same two ages. Success at 65 falls to 85.6%. Success at 40 falls to 60.9%. 65 remains higher. The title's ranking is inside this calibration and survived a 2-point equity-premium cut. It is still not a historical ranking.

Q3 was not run. A 50-year retirement has two independent 50-year windows in a century of tape, not a probability. Overlapping windows would overstate n. We did not quote a historical 4%-at-40 success rate. If 1966 or 1968 as a 50-year start failed this rule, that would be a veto, and we did not look. The takeaway stays simulated.

Q4: 5 bps, tax off. What would falsify: success at 40 greater than or equal to success at 65 on the base run, or the mu-shave flipping that order. Neither happened.

The rest of the age table, in one place

Age 50, 40-year horizon: 83.5% success, median $1,247,069, p95 $14,360,834. Age 55, 35 years: 86.2%, median $1,143,665. Age 60, 30 years: 89.6%, median $1,083,603. The slope is not a cliff at 40. It is a decade-by-decade tax on the same rate.

Notice that the 30-year B4_RETIREE cone (age 65, 30 years to 95) printed 89.6% at rigid 4%. The horizon table's age-65 row is 25 years to 90 and prints 93.5%. Those are not the same horizon. Mixing them in one sentence would be a methods hole. This page keeps them separate. The cone is a thirty-year life from 65. The age table is a life that ends at 90.

p95 explodes as the horizon lengthens because survivors have more time: $24,476,518 at 40 versus $6,528,822 at 65. That right tail is not a reason to call 4% at 40 safe. It is the other face of the same extra years that produced the extra ruin. Horizon risk is both tails.

See a long horizon on a calculator

The 4% rule calculator and retire at 40 are the matching tools. If the horizon slider does not move the success rate, the tool is not doing this experiment. This freeze moved it from 93.5% at 65 to 78.7% at 40 with nothing else changing except the number of years the $40,000 had to be paid. That is the smallest honest test of the title. Adding a job, a contribution, or a band would have been a different treatment. The card did not allow it. A calculator that silently adds those things while keeping the 4% label is not this experiment, and its 90% is not this 78.7%.

Write the sentence you can defend: rigid 4% from one million fails more often over fifty years than over twenty-five, on this engine, after a 2-point equity shave still, and not as a historical probability. Everything else is a different paper.

What this page refuses to say

It refuses to say 4% is safe at 65. 6.5% of those 25-year lives still fail. It refuses to say 4% is reckless at 40. 78.7% still last. It refuses to say early retirees should spend 3%. It refuses to merge this table with the August GBM table and call the blend a replication. It refuses to quote overlapping historical windows as n = 50. Q3 was not run. Independent 50-year windows in a century of tape are two, maybe three. That is not a probability.

It refuses to let p95 at 40 ($24,476,518) talk over the ruin mass. Extra years raise both tails. Horizon risk is the name for that pair, not for a feeling about youth. It refuses a flexible-rule 100% as a rescue of 4% at 40. A band that cuts spending is a different treatment. The card froze the rate so the horizon could be the only moving part. Un-freezing it after seeing 78.7% would have been search.

It refuses a personal retirement-age recommendation. Age 40 on this table is a 50-year sentence from a million, not a 32-year-old with a job and a contribution. The coast article is the contribution-stop version. Do not bolt them into one lifestyle claim.

Q2 survived. That lets the title keep the ranking inside this calibration. It does not let the title drop the scope sentence. The scope sentence stays in the takeaways: not a historical claim, pre-tax, 5 bps on.

Replication spec

Five populations, one seed. Each: $1,000,000 already retired, rigid 4%, 80/15/5, 1,000 paths. Only projectionYears changes: 50, 40, 35, 30, 25. Success at 40 is 78.7% (ruin 21.3%, median $1,574,241, p95 $24,476,518). At 65: 93.5% (ruin 6.5%, median $1,021,875, p95 $6,528,822). Gap: 14.8 points. Pairing is by seed, not by shared month index across different lengths.

Q2: stock mu −0.02, ages 40 and 65 only. 60.9% versus 85.6%. Medians $263,756 versus $546,258. Order on success survives. The title is allowed to keep "riskier at 40" as a Q1-plus-Q2 statement inside this calibration. It is not allowed to say historical, and this page does not.

Age 50 row: 83.5% success, $1,247,069 median. Age 55: 86.2%, $1,143,665. Age 60: 89.6%, $1,083,603. p5 is zero on every row. CSV: /data/studies/4-percent-rule-riskier-at-40-than-65.csv.

Do not confuse this 25-year age-65 row with the 30-year B4_RETIREE cone. Different horizons. Different endings. Same rate. Same seed. That is allowed. Mixing the success rates in one sentence is not.

Last recap: one million dollars, rigid 4%, five horizons to age 90. Success 78.7% at 40 and 93.5% at 65. Q2: 60.9% and 85.6%. Order survives. Not a historical probability. Not a safe stamp at 65. Not a reckless stamp at 40. Horizon risk is extra years for the same rate, both tails included, p95 at 40 $24,476,518 and ruin still 21.3%. New slug. Different generator than the August age table. Do not average them.

If a planner uses 4% at 40 because a 30-year paper said 4%, they changed the homework. This freeze is the changed homework, scored. Intermediate ages are on the CSV so nobody has to invent a cliff: 50, 55, and 60 sit on the slope between 78.7% and 93.5%. The slope is the finding. A cliff would have been a nicer headline and a worse experiment. The CSV is the table. This page is the caption. Do not invent a cliff. Do not average this slope with the August GBM table. Read the freeze file. Seed 20260622.

Notes. B4_HORIZON: $1,000,000 already retired, rigid 4%, 80/15/5, seed 20260622, 1,000 paths, horizons 50/40/35/30/25 years (ages 40–65 to 90). Q1 run. Q2 run on ages 40 and 65 with stock mu −0.02; order survived. Q3 not run (not a historical claim). Q4: 5 bps, tax off. CSV: /data/studies/4-percent-rule-riskier-at-40-than-65.csv. Card dated 2026-08-28. Not a rewrite of withdrawal-rates-by-retirement-age. This article is educational analysis, not investment advice, and does not recommend any security or strategy.