TL;DR: the answer box
- The usual bucket plan did not make the money last longer. At 4% a year, three buckets with the usual refill rule ran out in 4.0% of 749 start dates. A 60/40 rebalanced yearly ran out in 1.9%. At 5%: 28.8% against 22.2%.
- Of the bucket choices I tested, the refill rule moved the result most. Refill cash from stocks after an up year: 4.0%. Refill every year: 13.6%. Keep every bucket full: 15.1%. Never refill from stocks: 2.0%.
- The up-year test beat all 5 coin flips. A coin-flip refill, selling stocks just as often, ran out in 8.0% on average at 4%.
- More cash meant more failures. One year of cash in bucket one: 2.0%. Two years: 4.0%. Three years: 6.1%.
- No smoother ride either. The usual bucket plan's typical worst fall was 31.3%. The 60/40's was 26.1%.
What is the bucket strategy for retirement?
You retire with $1,000,000. You need an income from it for 30 years. The bucket strategy says: split the money by when you will spend it.
- Bucket 1, cash: the next 2 years of spending, in T-bills (short-term US government loans that pay a little interest).
- Bucket 2, bonds: years three to ten, so 8 years of spending, in 10-year Treasuries.
- Bucket 3, stocks: everything else, in the S&P 500.
The idea feels safe. When stocks crash, you live off cash and bonds and wait. That is the pitch: you never have to sell stocks at the bottom.
Every adviser page explains the buckets. Almost none tests them on history. So I did.
Think of the cash bucket as a fire extinguisher. You want one in the kitchen. But you do not want to keep buying new ones with money taken from the part of the house that is growing. Hold that picture. It explains most of what follows.
What I compared it with. A 60/40 portfolio: 60% stocks, 40% bonds, one pot. Each year it is rebalanced, which means reset back to 60/40. Your income sells a little of both.
At 4% a year the buckets start at 8% cash, 32% bonds, 60% stocks. At 5% a year the cash and bonds buckets are bigger, so it is 10% cash, 40% bonds, 50% stocks.
How I tested the retirement bucket strategy
The money. $1,000,000 at retirement. You take 4% of it a year (and, in a second run, 5%), paid monthly. Every 12 months the payment is raised with inflation (the rise in prices, measured by the CPI). Spending never changes because markets fell.
The start dates. Every month-end from February 1934 to June 1996: 749 retirements, each run for the full 30 years. 1934 is where the data starts, because it is the first year with a published T-bill rate.
These start dates overlap a lot. 92 years of data hold only about three separate 30-year retirements. So read every share below as "how often the starting month was a bad one", not as a probability.
The data. Monthly, January 1934 to June 2026. Stocks: the S&P 500 with dividends reinvested, from month-end index closes and Robert Shiller's dataset. Bonds: Shiller's 10-year Treasury series. Cash: the 3-month T-bill rate from FRED. Inflation: CPI.
The fills. Each yearly review is decided at a month-end close. Monthly data only has closes, so the trade fills at the next month-end close, one month later. That is not quite Execution rule v2 (decide at the close, fill at the next open). So I also ran the plans on real ETFs under Execution rule v2, further down. Everything runs through the same engine, so the same code scores every plan.
The plans I ran:
- 60/40, rebalanced yearly.
- Same mix, rebalanced yearly: the buckets' own day-one mix, reset every year. No buckets, just the mix.
- Buckets, usual rule: refill cash from stocks if stocks rose over the last 12 months. Otherwise refill from bonds.
- Buckets, keep every bucket full: as above, and after an up year also top bonds back up from stocks.
- Buckets, refill every year: refill cash from stocks at every review, up year or down.
- Buckets, never refill from stocks: spend cash, then bonds. Sell stocks only once the bonds are gone.
- Two yardsticks: 100% stocks, never rebalanced, and 100% T-bills.
The words I use:
- Ran out: the account could not pay a monthly payment in full before year 30.
- Worst fall (drawdown): the biggest drop from a high point. "Typical" means the median, the middle value across the 749 start dates.
- Money left: the balance at year 30 in retirement-day dollars (after inflation), as a share of the starting $1,000,000.
Does the bucket strategy make your money last longer?
Not with the usual refill rule. In this history, the plain portfolio lasted longer.
At 4% a year, the 60/40 lasted the full 30 years in 98.1% of start dates. The usual bucket plan lasted in 96.0%. Keeping every bucket full lasted in only 84.9%.
Flip it around and count the failures.
The usual bucket plan ran dry in 30 of 749 start dates. The 60/40 ran dry in 14. Same money, same income, same market history.
Here is everything on one page.
The bucket plan did end richer in a typical case. It left 235% of the start, against 171% for the 60/40. That is the trade it makes. The bond bucket gets spent and is not refilled, so the account drifts toward stocks. Great in a good decade. Worse when the bad decade comes early.
At 4%, the plain mix beat the buckets. The buckets' own day-one mix, simply rebalanced every year, ran out in just 0.5% at 4%, against 4.0% for the usual buckets and 1.9% for the 60/40. Same day-one mix. No buckets. Fewer failures. At 5% it lost that edge: 26.8%, more than the 60/40's 22.2%.
Michael Kitces tested one retirement, a 1966 start, and found a bucket plan and a rebalanced portfolio came out almost the same. This test is wider: 749 start months, two income levels and four refill rules. At 4%, all four refill rules ran out more often than the 60/40, though never refilling only just: 2.0% against 1.9%.
When did the bucket strategy run out?
At 4%, almost all of it happened to one generation.
Every 4% failure of the usual bucket plan began between 1964-04 and 1969-04. The 60/40's failures began between 1964-12 and 1968-10. Those retirees met the inflation of the 1970s early. Their income was being raised fastest just as stocks and bonds both struggled.
So the 30 failures are one bad decade, seen from 30 starting months. That is the honest size of the sample.
Here is one of them: a retirement at the end of 1965.
Keeping every bucket full ran out in 1990-02, after 24.2 years. The usual bucket plan ran out in 1994-01, after 28.1 years. The 60/40 ran out in 1994-04, after 28.3 years. Only 100% stocks reached year 30, with 1% of the start left, after a 42.8% fall on the way.
Does the bucket strategy protect you in a crash?
Mostly not. It fell more than the 60/40 in 3 of the 4 crashes.
This is the claim you hear most. The cash bucket lets you skip selling stocks in a downturn, so the plan is safer. Half true. You do avoid selling stocks for a while. But the account still falls.
I took four big stock crashes in the data, S&P 500 falls of 42.7% to 51.0%, and measured each account from the market's high to its low, income included. Each crash is read from a retiree a few years in.
The usual bucket plan fell less than the 60/40 in 1 of the 4 crashes (1973 to 1974: -31.6% against -33.4%). It fell more in the other 3. In 2007 to 2009 it fell -33.3% against -28.2%.
Why? Look at how the rule works. The usual plan spends its bonds into cash and refills cash from stocks only after an up year. So between refills nothing pulls it back to a fixed mix, the way a yearly reset pulls the 60/40 back to 60% stocks. In 2007 to 2009 the plan that never refills from stocks fell -52.5%, the most of the bucket plans and almost as much as 100% stocks at -53.2%.
Across all 749 start dates, the pattern holds.
At 4%, the usual bucket plan's typical worst fall was 31.3%, deeper than the 60/40's 26.1%. The two bucket plans that fell least (refill every year at 25.8%, keep every bucket full at 27.6%) are the two that ran out most. At the far end, 100% stocks ran out least, 0.1%, and fell the most, 42.8%.
If you want a smooth ride, look at the falls, not the bucket names. The label changes nothing. The refill rule changes what you own.
Which refill rule is best?
For making the money last, never refilling from stocks ranked first of the four.
Back to the fire extinguisher. Every refill from stocks moves money out of the part that grows fastest and parks it at the T-bill rate. Keep every bucket full, and you do that every good year, for 30 years.
Christine Benz of Morningstar warns against exactly this. In Morningstar's retirement bucket basics she says retirees should not be "constantly moving money from bucket 3 (stocks) to 2 (bonds) to 1 (cash)". The numbers agree with her. The money left tells the same story: the usual rule left 235%, keeping every bucket full 184%, never refilling 327%.
Is the up-year test doing anything? I swapped it for a coin flip that sells stocks to refill cash just as often as the real rule did (77.4% of reviews at 4%). Five tries, every start date. The coin ran out in 8.0% on average at 4% (5.7% to 11.6%). The real rule beat all 5 tries, at both 4% and 5%.
So waiting for an up year before selling stocks does help. It just helps less than not selling stocks to refill cash at all. Never refilling ran out in 2.0% at 4%. The price: a typical worst fall of 42.1%.
Here is the usual rule on real trades. I ran it on SPY (an S&P 500 fund) and IEF (a Treasury fund) from July 2016, with real daily prices and trading costs.
There were 9 yearly reviews. At 8 of them, stocks were up over 12 months, so the rule sold stocks to refill cash. At 1, in 2022, stocks were down, so cash came from bonds and stocks were left alone. Each decision is made at the July month-end close and filled at the next day's open.
How many years of cash should be in the first bucket?
Of the three sizes I tested, one year ran out least, and fell the deepest.
With the usual refill rule at 4%, one year of cash ran out in 2.0% of start dates. Two years: 4.0%. Three years: 6.1%. The typical worst fall went the other way: 33.5%, 31.3%, 28.8%.
Cash is a comfort. In this test, comfort cost money. 100% T-bills is the extreme case: it never fell, and it ran out in 70.1% of start dates at 4%.
What happens at a 5% withdrawal rate?
Every plan got much weaker, and the order barely changed.
At 5% a year the usual bucket plan ran out in 28.8% of start dates, the 60/40 in 22.2%. Keeping every bucket full ran out in 44.6%. Never refilling from stocks held up best of the bucket plans, at 18.7%. Raising the income from 4% to 5% raised the failure rate for every plan I tested.
Did real ETF trades match the index math?
Yes, on money left and on the order of the plans. Monthly index data can flatter a test. So I ran six plans on SPY and IEF with real daily opens and costs on every share, under Execution rule v2, over the ten years from July 2016.
The two differ by at most 1.5 percentage points and put the plans in the same order. The usual bucket plan paid $127 in trading costs over ten years. Daily data does show deeper falls, because a crash that recovers inside a month (March 2020) is invisible at month-end. Ten years cannot show a 30-year failure rate. This checks the arithmetic, not the verdict.
Does the S&P 500 really average 10% a year?
Yes, before inflation. Your spending money grew less.
From January 1934 to June 2026, with dividends reinvested, the S&P 500 grew 11.1% a year. Its worst month-end fall was 51.0%. After inflation it grew 7.2% a year. Ten-year Treasuries grew 4.8% a year, with a worst fall of 25.2%. A retiree spends real dollars, so the 7.2% is the one that shows what the money could buy. It is what this history paid after inflation, not a forecast.
The verdict, and the honest limits
Where the bucket story is right. An all-cash plan was not safe over 30 years. T-bills alone, paying the same rising income, ran out in 70.1% of start dates at 4%, and each extra year of cash I tested, from one to three, added failures. And waiting for an up year before selling stocks beat random timing in all 5 tries.
Where it is wrong. The usual bucket plan did not make the money last longer than one rebalanced portfolio: 4.0% against 1.9%. Its typical worst fall was deeper too: 31.3% against 26.1%. The most careful-sounding version, keeping every bucket full, was the worst: 15.1%.
The limits, in plain words:
- The 749 start dates overlap. At 4%, every failure of the usual bucket plan began between 1964 and 1969. At 5% the failures spread from 1935 to 1973.
- US stocks, 10-year Treasuries and T-bills only. Monthly data. No taxes, and no trading costs on the 92-year run.
- Spending never changes. Real retirees cut back after a bad year, and that alone changes these numbers.
- One bond bucket size (8 years) and one horizon (30 years).
- I did not record each plan's stock share on the crash dates. So I measured that the bucket plan fell more in 3 of the 4 crashes, not why. The drift toward stocks is my reading of its rule, not something this test proves.
- This measures withdrawal plans on history. It is not financial advice.
What this means for you
- "It feels safer" is not evidence. In 749 start dates the usual bucket plan ran out more often than a 60/40 rebalanced once a year: 4.0% against 1.9% at 4%.
- If you use buckets, the refill rule is the real decision. Refilling only after an up year ran out in 4.0%. Keeping the bond bucket topped up from stocks ran out in 15.1%. Never refilling from stocks ran out in 2.0%, with a deeper fall.
- Test a smaller cash bucket. One year of spending ran out in 2.0% of start dates. Three years ran out in 6.1%.
- Decide what you are protecting against. No plan was best at both. 100% stocks ran out least (0.1%) and fell most (42.8%).
- Test your own rule before you trust it. A withdrawal plan is a strategy you only get to run once. I entered the markets in 1998, at the dot-com peak. Your start date is the one input you do not get to pick. If you build your own rules, AlgoChef is the tool I use to check whether a backtest holds up.
The same portfolio calculator runs the tactical asset allocation study and the dollar cost averaging study.
Buckets are a way to label money. The refill rule is the strategy.
Reference tables
Every plan, 749 start dates, 30 years each. Money left is the typical (median) balance at year 30 in retirement-day dollars, as a share of the start. Turnover is the money traded each year as a share of the balance.
| Plan | Ran out at 4% | Ran out at 5% | Typical worst fall (4%) | Money left (4%) | Yearly turnover (4%) |
|---|---|---|---|---|---|
| 60/40, rebalanced yearly | 1.9% | 22.2% | 26.1% | 171% | 5.2% |
| Same mix, rebalanced yearly | 0.5% | 26.8% | 26.5% | 176% | 4.0% |
| Buckets, refill after an up year | 4.0% | 28.8% | 31.3% | 235% | 1.6% |
| Buckets, keep every bucket full | 15.1% | 44.6% | 27.6% | 184% | 2.4% |
| Buckets, refill every year | 13.6% | 42.6% | 25.8% | 178% | 1.8% |
| Buckets, never refill from stocks | 2.0% | 18.7% | 42.1% | 327% | 1.4% |
| 100% stocks | 0.1% | 14.8% | 42.8% | 413% | 1.2% |
| 100% T-bills | 70.1% | 95.9% | 0.0% | 0% | 0.0% |
The four crashes. Account balance change from the S&P 500's month-end high to its low, at 4% a year, income included.
| Crash | Retired | S&P 500 | 60/40 | Buckets, usual | Buckets, never refill | 100% stocks |
|---|---|---|---|---|---|---|
| 1937 to 1938 | 1937-02 | -49.7% | -31.7% | -32.7% | -32.7% | -52.4% |
| 1973 to 1974 | 1968-12 | -42.7% | -33.4% | -31.6% | -37.5% | -48.3% |
| 2000 to 2002 | 1995-12 | -44.7% | -24.3% | -31.8% | -39.9% | -47.7% |
| 2007 to 2009 | 1995-12 | -51.0% | -28.2% | -33.3% | -52.5% | -53.2% |
Everything here is research, not investment advice. It is a backtest on historical data with the assumptions stated above. It is not a recommendation to buy or sell anything, and past results do not predict future ones.
I test one strategy everyone repeats, properly, and leave the parts that failed in. StatOasis.com/Overfit
