The bucket debate argues whether to hold cash. The useful question is how much. So I ran buffers of zero to five years through every 50-year retirement since 1928: the answer is a narrow window, two to three years, and past it the cash itself becomes the risk.
The cash buffer debate always gets argued as a yes-or-no question: hold a few years of cash so a crash can't force you to sell, or stay fully invested because cash drags. The famous research answer, from Michael Kitces, is that reserve strategies mostly don't work. My calculator uses a buffer anyway, so I owed it the harder question nobody runs: not whether, but how much. I tested every buffer size from zero to five years against every full 50-year retirement since 1928, using the same engine that runs your numbers on the calculator.
| Buffer size | 50-yr windows survived | Median ending | Worst 10% ending | Max safe rate |
|---|---|---|---|---|
| None | 47/48 | 8.8× | 1.15× | 3.45% |
| 1 year | 47/48 | 8.2× | 1.14× | 3.45% |
| 2 years | 48/48 | 7.9× | 1.32× | 3.50% |
| 2.5 years | 48/48 | 7.7× | 1.25× | 3.55% |
| 3 years | 48/48 | 7.5× | 1.06× | 3.65% |
| 4 years | 48/48 | 6.8× | 0.68× | 3.55% |
| 5 years | 46/48 | 6.1× | 0.33× | 3.40% |
All-equity inflation-adjusted S&P 500 returns 1928–2024, 3.5% constant real-dollar withdrawals over 50 years, buffer carved out of the same starting wealth and measured in years of total spending. Endings are multiples of starting wealth; max safe rate is the highest rate at which every window survived. “Worst 10% ending” is the 10th-percentile outcome — one in ten historical retirements ended at or below it.
Both lines fall as the buffer grows — that's the cash drag, in multiples of starting wealth. The shaded band is where the survival record (the row of counts, out of 48) is perfect. Past three years the worst outcomes collapse: by five, the unluckiest retirements end with a third of what they started with.
Three lines in that table tell the whole story. Two years is the minimum that achieves a perfect record: it's exactly the size that rescues the one failing start, 1929, that kills the bare portfolio at 3.5%. Three years buys the most rate headroom, lifting the never-failed withdrawal rate to 3.65%. And five years is worse than holding no buffer at all: two new failures appear, 1966 and 1969, because so much of the portfolio is sitting out of the market that the drag itself becomes the danger, and the unluckiest tenth of retirees end with a third of their starting wealth instead of a multiple of it. Meanwhile the median column falls with every year of cash you add, which is the drag critics talk about, measured: each extra year of buffer costs you real legacy in the typical outcome. So the sweet spot is narrow and real: two to three years. Anywhere inside that window is fine; the data doesn't crown a single point. Two years leaves the most invested, three buys the most rate headroom, and the middle simply splits the difference.
The best-known work here argues against me, and it deserves a straight summary. Kitces's analysis found that cash reserve strategies mostly reduce long-term returns without improving outcomes, since ordinary rebalancing already avoids selling stocks in down markets, and a follow-up academic study by Estrada across 21 countries and 115 years reached the same conclusion: static allocations beat bucket strategies. My table agrees with the mechanism. The drag is real and shows up in every median, big buffers are strictly harmful, and at a 4% rate over 50 years a buffer of any size adds exactly one surviving window, because when the withdrawal rate itself is the problem, cash can't fix it. Where my result differs is scope: a small buffer, two to three years, paired with a conservative rate and a refill rule that only tops the cash back up after strong market years, buys measurable worst-case protection, the 1929 fix and the perfect 30-year record, at a known price of roughly one multiple of median legacy. That's not a return enhancer. It's insurance, priced.
If the table had to be one sentence: hold two years if maximizing what you leave behind matters most, three if you want the most withdrawal-rate headroom, and never five. Spending here means total spending, health insurance included, since that's what you'd actually draw in a crash. The sequence-of-returns page explains the mechanism this protects against, the 50-year backtest covers the withdrawal-rate side of the same data, and the methodology is the calculator's own runHistoricalSequence function: full windows only, no fees, no taxes, spend from cash in negative years, refill after years above +10%.
Not financial advice. Consult a fee-only fiduciary CFP for personalized guidance. Tax figures use 2026 brackets.