Bond Tent, Cash Bucket or Guardrails: Which Handles Sequence-of-Returns Risk Best? (2026)

Bond tent, cash bucket or guardrails: which handles sequence-of-returns risk best?

Key numbers

10,000 quasi-Monte Carlo paths per arm, identical market draws in every arm; JPMorgan LTCMA 2026 capital-market assumptions (as of 2025-11-01); all dollars in today’s dollars; pre-tax, no Social Security. “Success” means the portfolio never reached $0 before 95. Deciles rank the 10,000 paths by the first-decade real return of the 60/40 portfolio, as in the site’s sequence-of-returns study.

Run the $1M / $40,000 / 65-to-95 plan in the free planner →

Opens the Monte Carlo planner prefilled with this study’s plan — retire at 65, plan to 95, $1,000,000, $40,000 a year, 45/15/40 US equity / international / bonds. The link sets the plan, not the defence: choose Guardrails or Bucket (cash reserve) under Withdrawal Rule, or add allocation periods for a glide path, to see each arm for your own numbers.

What is being compared

The control is the site’s standard plan: 45/15/40 US equity / international equity / bonds, rebalanced every month, spending $35,000 / $40,000 / $45,000 a year in today’s dollars from 65 to 95 with no Social Security or pension. Each defence is one existing input of the engine, run on the same 10,000 market paths:

• Bond tent — changes what you hold. A multi-period glide path that starts at 40/60 equity/bonds at 65 and steps up every few years to reach 60/40 (tent to 60/40) or 70/30 (tent to 70/30) at 75, then stays there. The two versions bracket the control: over the 30 years the tent to 60/40 averages 55.8% equity, the tent to 70/30 63.7%, the control 60%. Spending is level in both.
• Cash bucket — changes what you sell. The engine’s two-bucket rule: 3 or 5 years of spending carved out of the portfolio into cash at retirement, drawn on first, refilled to target at year end only after a year in which the invested sleeve did not lose money; the sleeve stays at 60/40, rebalanced annually. Spending is level. Because a bucket portfolio holds cash the control does not, each bucket arm is also paired with a matched-cash control: level spending from a portfolio that starts with the same 12.0% (3 years) or 20.0% (5 years) in cash, the rest 60/40, rebalanced annually — the construction documented alongside the engine’s bucket rule and used by the site’s bucket study, whose cells this study reproduces exactly.
• Guardrails — change what you spend. Guyton-Klinger on the 60/40 control: at each retirement anniversary the current withdrawal rate (this year’s real spending ÷ real portfolio) is compared with the rate set at 65; more than 20% above it and spending is cut 10%, more than 20% below and it is raised 10%; no cut is applied inside the last 15 years. Spending is otherwise level in real terms. The engine reports what each path actually spent.

Every arm starts from the same $1,000,000 on the same paths. The headline statistic for each arm is the paired difference against the control — this arm’s wealth at 95 minus the control’s on the same path — summarised by its median, its 10th–90th percentile band, the share of paths on which the arm ends higher, and the paths only one of the two survives. For the guardrails arm the paired figure must be read next to its realised spending: a higher balance bought with lower withdrawals is not a free improvement. Every statistic is then cut by decile of first-decade return: paths are ranked by the real annual return the 60/40 portfolio earned over its first ten years with no withdrawals (worst decile mean -1.67% a year, best decile 9.60%), which is the sequence-of-returns cut used in the site’s sequence-of-returns study.

Results: every arm against the control

Dollar columns are wealth at 95 in today’s dollars; “Spending” is the 10th percentile and median across paths of each path’s average annual real spending over the 30 years — the plan itself for every arm except guardrails. The matched-cash controls and the annually rebalanced control are included as arms so their differences from the control are measured, not assumed.

$35,000 a year (3.5% of $1,000,000)

$40,000 a year (4.0% of $1,000,000)

$45,000 a year (4.5% of $1,000,000)

“Paired median Δ” is the median over 10,000 paths of (arm wealth − control wealth) at 95; “Ends higher” is the share of those paths with a positive difference. Paths where both arms end at $0 are ties, which is why a paired median can be exactly $0 at higher spending: the middle of the per-path distribution sits inside that tie mass. The annually rebalanced control differs from the monthly one by −0.2 pp of success and a paired median of $0 at $40,000; that is the size of the cadence effect carried by the bucket and matched-cash arms, which the engine rebalances annually.

The cash bucket against its matched-cash control

The fair test of the bucket is against a level-spending portfolio that starts with the same cash. Every row below is one of the site’s bucket-study cells, re-run here on the same engine and reproduced to the cent: at $40,000 with a 3-year reserve, 85.2% vs 87.0% success, paired median +$10,705, bucket higher in 57.4% of paths there and here.

With the cash matched, the bucket runs out more often at every spending level and reserve size (by 0.8 pp to 2.5 pp for 3 years, 1.5 pp to 5.9 pp for 5), and does worse on exactly the paths it is meant for: in the worst decile at 4.0% spending, 22.0% of 3-year-bucket paths reached 95 against 29.1% of matched-cash paths. The reserve is a set number of years of spending, so a shrinking portfolio carries proportionally more cash and every refill after a loss sells the invested sleeve to buy it, while the matched control rebalances back into what fell. What the bucket gains — a higher median where markets are kind, from carrying proportionally less cash as the portfolio grows — is the opposite of sequence-risk protection.

The sequence-risk cut: worst-decile against best-decile first decades

Each decile is 1,000 paths. The worst decile’s first ten years earned a mean real return of -1.67% a year on the 60/40 portfolio; the best decile’s 9.60%. Paired columns are against the control on the same paths. “Median spending” is the median across the decile’s paths of each path’s 30-year average annual real spending.

$35,000 a year (3.5%)

$40,000 a year (4.0%)

$45,000 a year (4.5%)

Three patterns hold at every spending level. Guardrails reach 95 on at least 98.7% of worst-decile paths, and on every one of those paths spending was cut at least once; the median worst-decile path under guardrails lives on 28.9% to 36.0% less than its plan. The bond tent lifts worst-decile success by +5.3 pp to +15.2 pp (tent to 60/40) and +7.4 pp to +11.5 pp (tent to 70/30) while giving up best-decile wealth in every case. The bucket is below its matched-cash control in the worst decile in every case.

Success by decile of first-decade return, $40,000 a year

Almost all of the control’s failures sit in the bottom three deciles; that is what sequence-of-returns risk looks like on a 4% plan. The tent moves the bottom rows, the bucket lowers them, and guardrails flatten the whole column by moving the risk into the spending column instead.

What guardrails cost in spending, $40,000 a year

Every path starts at $40,000. The table shows the distribution across all 10,000 paths of that year’s real spending, and the share of paths spending below plan that year. Over the whole retirement 60.4% of paths took at least one cut and 56.7% at least one raise; 28.2% had been cut by 70 and 49.5% by 75; 44.1% were still below plan at 94. The median path’s lowest single year was $36,000; the 10th-percentile path’s $23,620.

Real spending in the year the retiree is that age, today’s dollars. The engine applies a 10% cut when the withdrawal rate exceeds 20% above its starting level (and more than 15 years remain) and a 10% raise when it falls more than 20% below; adjustments compound, so a path cut twice spends $32,400.

The gap over time, $40,000 a year

Median per-path difference against the control (this arm minus control, today’s dollars) at six ages, with the share of paths on which the arm is ahead. The tent’s gap is widest around 75, when its ten bond-heavy years are complete and its equity has only just caught up; the guardrails gap grows as cuts compound on the paths that need them.

What each defence does, and what it costs

Bond tent: changes what you hold

Sequence risk bites when a large equity share falls in the first years and level withdrawals are taken from the depressed balance. A tent shrinks the equity share exactly then. At $40,000 the tent to 60/40 turned 217 control failures into successes and lost 61 the other way; the tent to 70/30, 194 against 109. Worst-decile success rose from 28.9% to 39.3% and 39.0%. The cost is upside: the tent holds less equity in the years when the best decile makes its money, so it ends $447,470 and $225,178 lower at the best-decile paired median, and across all paths the tent to 60/40 ends $53,985 less at the median, higher on only 26.6% of paths. The tent to 70/30 holds more equity than the control on average and ends the same at the median (higher on 43.3%, lower on 45.8%), which is the fairer comparison of shape alone: the same or more risk overall, arranged so that less of it is taken early.

Cash bucket: changes what you sell

The bucket leaves the invested sleeve alone after a losing year and spends cash instead. That is a real mechanism — it rescued 21 paths at $40,000 that the matched-cash control lost — but the matched control lost 198 paths the bucket kept, because the bucket never rebalances back into what fell and its refills after a loss sell the sleeve to rebuild cash. The reserve also drags: against the cash-free control the 3-year bucket ends $135,179 less at the median and the 5-year bucket $262,558 less, most of it the return given up on 12.0% and 20.0% of the portfolio held in cash. A bigger bucket is worse on both counts (83.1% success against 86.8% matched; 16.7% against 29.4% in the worst decile). The value a visible cash reserve has in helping a person stay invested through a bear market is outside these numbers.

Guardrails: change what you spend

Guardrails act on the mechanism itself: when the withdrawal rate has drifted 20% above its starting level, the next year’s spending is 10% lower, and the cut compounds into every later year until a raise reverses it. On the worst-decile paths at $40,000 every path was cut, the median path averaged $26,920 a year (32.7% below plan), its lowest year was $21,258, and it spent 26 of 30 years below plan — and 99.8% of those paths reached 95 with money left, a median $457,962 against the control’s $0. On the best-decile paths the same rule raised the median path to $60,888 a year and left $907,588 less at 95 than the control, because it spent it. Success under guardrails is therefore not comparable with success under a level rule: the 99.9% is bought with a 10th-percentile average spending of $29,089 against a planned $40,000. The site’s withdrawal-strategy study examines that trade on its own terms.

What it means

• Name the risk before choosing the defence. If the risk is running out, only a spending rule removes it, and it does so by making spending the thing that varies. If the risk is a cut to spending, the tent is the only one of the three that lowers the chance of ruin without ever touching the withdrawal, and it costs median wealth.
• A bond tent is a modest, real defence. +0.8 pp to +1.8 pp of success overall and +5.3 pp to +15.2 pp in the worst decile for the tent to 60/40 across the three spending levels; the price is paid in the good decades.
• A cash bucket is not a sequence-risk defence. Below its matched-cash control at every spending level, every reserve size, and in the worst decile every time.
• Guardrails convert ruin risk into spending risk. At least 99.0% success overall and 98.7% in the worst decile at every spending level, with the median worst-decile path living on 28.9% to 36.0% less than plan.
• The defences are not exclusive in practice, but they are in this engine. Its glide path ignores dynamic withdrawal rules, so a tent combined with guardrails could not be run; on the mechanism, the two address different halves of the problem and their effects would be expected to add.

CC-BY-4.0 — free for any use including republication and journalism, with attribution to QuantCalc Research.

Methodology

Assumptions and limitations

• The tent is a five-step staircase, not a smooth line. The contract runs at most five allocation periods, so equity rises in four steps between 65 and 75. A smoother glide path would sit between these results and the control’s.
• The three defences could not be combined. The engine’s glide path ignores the bucket and guardrail rules, so no arm mixes a tent with either; the bucket and guardrails were not combined with each other because the engine runs one withdrawal rule at a time.
• Rebalancing cadence differs by code path. The control, tents and guardrails rebalance monthly; the bucket’s growth sleeve and its matched-cash control rebalance annually. The annually rebalanced control is reported as an arm so the size of that effect is visible (−0.2 pp of success at $40,000).
• Guardrails as implemented. Symmetric ±20% bands around the initial withdrawal rate, ±10% steps at each anniversary, no cut inside the last 15 years, no separate inflation-freeze rule; other published variants would give different spending paths.
• Two buckets, not three. A cash reserve plus one invested sleeve at the plan mix; no separate bond bucket. The reserve earns the roster’s cash return (3.2% nominal).
• Deciles are defined on the 60/40 portfolio’s first-decade return, whichever arm is being examined, so every arm’s worst decile is the same 1,000 market paths.
• Parametric returns. Lognormal returns with the stated means, volatilities and correlations; no regime switching, fat tails or historical block resampling. The site’s sequence-of-returns study, on resampled history, finds a wider spread between its worst and best deciles.
• Pre-tax, no fees, no behaviour. No taxes, account types, fund expenses or trading costs; the control sells into every downturn as its rule requires; any value a defence has in helping a person stay the course is outside these numbers.
• Not advice. Educational research on three ways of arranging a retirement portfolio and its withdrawals; not financial, tax or legal advice.

Frequently asked questions

Bond tent, cash bucket or guardrails: which handles sequence-of-returns risk best?

They handle different things, and only one of them removes the risk of running out. For a 65-year-old with $1,000,000 spending $40,000 a year (4.0%) to 95 on a 60/40 plan, 10,000 identical market paths: the constant 60/40 control reached 95 with money left in 87.1% of paths, and in 28.9% of the paths whose first decade fell in the worst 10% of first-decade returns. Guyton-Klinger guardrails reached 95 in 99.9% of paths and 99.8% of the worst-decile paths, but on those worst-decile paths the median path averaged $26,920 a year instead of $40,000, its lowest year was $21,258, and it spent 26 of 30 years below plan. A bond tent (equity 40% at 65 rising to 60% or 70% by 75) lifted success to 88.7% and 88.0% overall and to 39.3% and 39.0% in the worst decile, while ending lower than the control in the best decile (paired median -$447,470 and -$225,178). A 3-year cash bucket reached 95 in 85.2% of paths against 87.0% for a level-spending portfolio holding the same cash, and 22.0% against 29.1% in the worst decile: it did not protect the retirement.

What does each defence actually change?

A bond tent changes what you hold: the portfolio starts bond-heavy and moves into equities over the first ten years, so a crash in the first years hits a smaller equity share. A cash bucket changes what you sell: spending is drawn from a cash reserve first and the invested portfolio is not sold after a losing year, but the amount spent never changes. Guardrails change what you spend: when the withdrawal rate drifts more than 20% above its starting level the year's spending is cut by 10%, and raised by 10% when it drifts more than 20% below. Sequence-of-returns risk is the risk that early losses plus level withdrawals exhaust the portfolio; a rule that reduces withdrawals after losses attacks that mechanism directly, which is why guardrails reached 95 on 99.8% of the worst-decile paths at 4.0% spending while the tent reached 39.3% to 39.0% and the bucket 22.0%.

How much spending do guardrails give up?

Across all 10,000 paths at $40,000 a year, 60.4% of paths took at least one cut and 56.7% at least one raise; the 10th-percentile path averaged $29,089 a year over 30 years, the median $40,000 and the 90th percentile $56,819. 28.2% of paths had been cut at least once by age 70 and 49.5% by 75; the median path's lowest single year was $36,000, the 10th percentile's $23,620. On the worst-decile paths every path was cut, the median path averaged $26,920 (32.7% below plan) and spent 26 of 30 years below plan; on the best-decile paths the median path averaged $60,888. Because a cut is never applied inside the last 15 years, late losses are absorbed by the portfolio, which is why 14 of 10,000 paths still ran out.

Why does the bond tent end lower at the median?

Because for its first ten years it holds less equity than the 60/40 control, and equities carry the higher expected return under the JPMorgan LTCMA 2026 assumptions. The tent to 60/40 averages 55.8% equity over the 30 years against 60% for the control, so it ends $53,985 less at the median and higher on only 26.6% of paths, with a paired median of -$447,470 in the best decile. The tent to 70/30 averages 63.7% equity, more than the control, and ends the same at the median (higher on 43.3% of paths, lower on 45.8%), but still -$225,178 in the best decile because the equity it adds arrives after the years in which the best decile made its money. What a tent buys is concentrated in the bad first decades: $0 and $0 at the worst-decile paired median at 4.0% spending, and at 3.5% spending, where more worst-decile paths survive, +$47,813 and +$51,176.

Why is the cash bucket compared with a matched-cash control as well as the 60/40 control?

Because a bucket portfolio holds cash the 60/40 plan does not, and the comparison is only fair if the level-spending side starts with the same amount. A 3-year reserve on $40,000 of spending is 12.0% of $1,000,000, so the matched control holds 60/40 scaled to 88.0% plus 12.0% cash, rebalanced annually — the construction documented alongside the engine's bucket rule and the baseline of the site's bucket study, whose cells this study reproduces exactly (85.2% vs 87.0% success, paired median +$10,705, higher in 57.4% of paths). Against the cash-free 60/40 control the bucket's paired median is -$135,179, most of which is the cost of holding 12.0% in cash rather than the withdrawal rule; against the matched control it is +$10,705. On the question this study asks the answer is the same either way: the bucket ran out more often than both controls at every spending level, and did worse than its matched control in the worst decile (22.0% vs 29.1% at 4.0%).

Are these numbers real dollars, and are taxes included?

All dollar figures are in today's dollars: the engine raises spending with 2.5% inflation and deflates every reported balance back to the retirement date. Nothing is after tax, there is no Social Security or pension income, and every arm except guardrails spends the same level real amount every year. Returns follow the JPMorgan LTCMA 2026 capital-market assumptions the site's planner uses, sampled with quasi-Monte Carlo, 10,000 paths per arm, identical draws in every arm.

Related research

Changelog

• v2026.1 (2026-09-08) — initial release. 27 arms (3.5% / 4.0% / 4.5% spending; two bond tents, two cash buckets with matched-cash controls, guardrails, two control cadences) on 10,000 paired paths; per-path paired statistics at 95 and by year; realised spending per path for guardrails; every statistic by decile of first-decade return.

Last updated 2026-09-08. Dataset license: CC-BY-4.0. QuantCalc is an independent retirement-planning research project. The JPMorgan LTCMA 2026 name identifies the published capital-market assumptions the engine uses; QuantCalc is not affiliated with, endorsed by, or sponsored by that firm, and all trademarks belong to their respective owners. Educational research, not financial, tax, or legal advice.

@misc{quantcalc2026bondtentcashbucketorguardrailswhichhandl,
  title  = {Bond Tent, Cash Bucket or Guardrails: Which Handles Sequence-of-Returns Risk Best? (2026)},
  author = {{QuantCalc Research}},
  year   = {2026},
  url    = {https://quantcalc.app/research/bond-tent-vs-bucket-vs-guardrails-2026/},
  note   = {Accessed <date>}
}