Cash Buffer in Retirement: One Year to Five, Tested

Key takeaways

• Across 86 rolling 30-year US retirements from 1900 to 2014, one year of cash beat two, three and five years under all three bucket rules Estrada tested, and five years pushed failure above 25%.
• Each year of spending held in bills instead of equities cost about 0.18 percentage points a year at the UK 1900-2014 spread, so a five-year buffer cost roughly 0.88 points a year.
• In Woerheide and Nanigian's 1926-2009 test, a one-year buffer beat a static allocation once in 140 comparisons; a four-year buffer won 13 times and lost 77.
• The best case for a buffer is a 2013 simulation where a one-year reserve lifted 30-year survival by roughly 6 points, and nearly all of that came from transaction costs.
• In the 2026 to 2027 tax year a £150,000 cash buffer at 3.75% owes £925 in tax at basic rate and £2,050 at higher rate outside an ISA, and exceeds the £120,000 FSCS limit at one bank.

How many years of cash in retirement: the evidence says one, and the tax rules say where

How many years of spending should sit in cash once you've stopped earning? One year, two, five? The guides disagree, and most give a number without a test behind it.

Here's what the tests say. In the longest one, Javier Estrada's 115-year study at IESE Business School, one year of withdrawals in bills beat two, three and five years under every bucket rule he tried. Five years pushed the failure rate of a 30-year US retirement from around 2% to over 25%. A second study, on US data from 1926 to 2009, found a one-year buffer beat a plain rebalanced portfolio once in 140 comparisons; a four-year buffer won 13 and lost 77. The one study that found a benefit found it for one year, mostly through saved dealing costs.

So the answer clusters tightly around one, and the interesting question moves. A cash buffer in retirement has a cost per year you can put a number on, and in the UK that cost depends on the wrapper. The cash bucket strategy guide covers whether buckets beat a rebalanced portfolio at all. This piece takes the size question and then the tax question, because the second decides what the first costs you.

What one year of cash costs before tax

Start with the spread between what cash pays and what equities pay, both after inflation. The 2026 UBS Global Investment Returns Yearbook puts US equities at 6.6% a year in real terms from 1900 to 2025, bonds at 1.6% and Treasury bills at 0.5%. That's 126 years of data and a gap of 6.1 percentage points between shares and bills. Estrada's dataset, 1900 to 2014, gives the UK 5.3% real on equities and 0.9% on bills, a gap of 4.4 points.

Now the arithmetic. At a 4% initial withdrawal rate, which every study here assumes, one year of spending is 4% of the starting portfolio. Five years is 20%. Held in bills rather than equities, each slice earns the bill return instead, so the drag on the whole portfolio is the slice multiplied by the spread.

Those are derived figures, and they compare cash against equities rather than against a 60/40 mix. A buffer carved out of bonds costs less. But the shape is what matters. The cost of a cash buffer in retirement is linear in the number of years, and at five years it's close to a full percentage point a year on a UK spread. That's the mechanism behind the failure rates in the next section, and it's why the cost never shows on a statement. It arrives as money you never had.

The 115-year test ranks one year first under all three bucket rules

Estrada's paper, dated December 2018, runs 86 overlapping 30-year retirements per country across 21 countries, from 1900-29 to 1985-2014. Each retiree withdraws 4% in year one, raises it with inflation, and holds two buckets: bills and stocks. Rule 1 spends from stocks if last year's real return was positive, rule 2 only if last year beat the long-run mean, and rule 3 only if the trailing five-year mean did.

The main results park two years in bills. Exhibit 4 reruns everything with one, three and five years, and that exhibit is the answer to this article's question.

The chart plots the third rule, the best of the three, against the two static portfolios Estrada uses as a yardstick. A static 70/30 and a static 50/50 of stocks and bills, rebalanced every year, each failed 1.2% of the time. A static 60/40 never failed. One year of cash under rule 3 matched the static 1.2%. Every extra year moved away from it. Estrada's own summary: "setting aside one year of withdrawals outperforms setting aside two, three, or five years of withdrawals for the three bucket rules discussed here."

Why does one year do so much better than five? Five years of spending is a fifth of everything you own, earning 0.9% real in the UK data while the rest earns 5.3%. And bucket rules move money one way, from stocks into cash. They avoid selling low. They never buy low, which is what annual rebalancing in the static portfolios does for free.

One qualification, because the US is the flattering sample. Across the 21-country average, a static 60/40 failed 30.7% of the time and Estrada's best bucket rule with two years of cash failed 30.1%. Globally the static advantage shows up in shortfall years and risk-adjusted measures rather than raw failure counts. The US table is the sharp version of the result, not the typical one.

A second dataset finds the damage grows with every added year

Walt Woerheide and David Nanigian took the opposite route in the Journal of Financial Planning in May 2012. They reran the methodology of Cooley, Hubbard and Walz, the Trinity study authors, on US returns from 1926 to 2009, with horizons of 15 to 30 years, withdrawal rates of 3% to 9%, five stock/bond mixes, and money-market buffers of one to four years. That gives 140 comparisons per buffer size, 560 in all, each a head-to-head against the same portfolio with no buffer.

The tally by size is the useful part for this question. A one-year buffer produced the better success rate 1 time and lost 32. Two years won 9 and lost 50. Three years won 12 and lost 67. Four years won 13 and lost 77. Their explanation is the same drag arithmetic: the return gap grows with the size of the buffer.

The withdrawal rate mattered as much as the size. At a 3% rate the buffer won 3 comparisons and never lost. At 4% it won 9 and lost 16. At 8% and 9% it never won at all. Their conclusion is that at 3% "it doesn't really matter whether one uses buffer zones", and that once equities enter the portfolio or the rate goes above 3%, a buffer of any size from one to four years is "more likely than not to leave the investor worse off". Which safe withdrawal rate is sustainable in the first place is a separate argument.

The strongest case for a buffer is one year, and it rests on dealing costs

One study finds a cash reserve helps, and it's worth reading properly because one of its authors, Harold Evensky, is the planner usually credited with inventing the approach. Shaun Pfeiffer, John Salter and Evensky published it in the Journal of Financial Planning in September 2013. They ran 1,000 Monte Carlo simulations of a 60/40 portfolio, assuming 8.75% nominal on stocks, 4.75% on bonds and 3.50% on cash, with 3% inflation. One version drew pro rata from stocks and bonds every month. The other held one year of spending in cash and refilled it when rebalancing, provided at least one asset had risen the year before.

With no taxes and no dealing costs, the reserve was slightly worse: 76.0% of plans survived 30 years at a 4% real withdrawal, against 76.9% without it. Add a $30 cost per trade and the picture flips: 75.7% with the reserve, 70.8% without. In a taxable account the reserve's survival rate was 66%, roughly 6 points ahead. The reason is plain: the pro-rata strategy ran up $21,600 in cumulative dealing costs by year 30, the reserve strategy $2,280, because it sold far less often.

So the benefit is real, but it's a benefit of trading less, not of holding cash. Two things follow. First, the authors tested six-month and two-year reserves too and report that the one-year version "generally" did best, so even the pro-buffer study lands on one year. Second, $30 a trade was a 2013 US brokerage assumption. A retiree paying nothing per trade, or drawing from a single automatically rebalanced fund, gets most of that saving with no cash at all. What's left is the behavioural argument, which Evensky's firm has made since 1985 and no controlled comparison has measured.

Joe Tomlinson's April 2020 reply to Estrada makes a related point. His Monte Carlo runs, at a 5% real arithmetic return on stocks with a 20% standard deviation and 1% on cash, found a two-year bucket and a static portfolio with the same starting mix "virtually equivalent". Year-to-year stock returns showed a correlation of -0.00069 from 1926 to 2019, so a rule keyed to last year's return has nothing to exploit. Note what that concedes. The bucket adds nothing, and the cash inside it is a lower equity weight wearing a different label.

Where a cash buffer in retirement sits decides what it costs after tax

Every study above ignores UK tax, and in a UK decumulation plan the wrapper changes the cost of a year of cash more than the choice between one year and two. All figures below are for the 2026 to 2027 tax year, as GOV.UK states them at the time of writing. They change, so the date matters.

Outside any wrapper, savings interest is income. The Personal Savings Allowance lets a basic-rate taxpayer receive £1,000 of interest tax-free, a higher-rate taxpayer £500, and an additional-rate taxpayer nothing. Above that, interest is taxed at 20%, 40% or 45%, with the basic band running from £12,571 to £50,270 and the higher band to £125,140. There's also the starting rate for savings: up to £5,000 of interest at 0%, in full only if other income stays under the £12,570 Personal Allowance, and gone at £17,570. It's easy to overlook and it matters most to someone whose only other income is a modest pension. The same annual thresholds are what make the personal allowance taper and the basic rate band so expensive to cross with a single large withdrawal. The same annual thresholds are what make the personal allowance taper and the basic rate band so expensive to cross with a single large withdrawal.

Put a number on it. Say you draw £30,000 a year and Bank Rate is 3.75%, where the Bank of England held it on 30 July 2026. A five-year buffer of £150,000 earns £5,625 a year. A basic-rate taxpayer pays £925 on that after the allowance, a higher-rate taxpayer £2,050. A one-year buffer of £30,000 earns £1,125, which costs the basic-rate taxpayer £25 and the higher-rate taxpayer £250. Savings accounts don't pay Bank Rate, so treat those as the shape of the bill rather than the bill.

Inflation takes its cut before tax does. CPI rose 2.9% in the 12 months to July 2026, on the ONS release of 19 August. Against a 3.75% Bank Rate that leaves about 0.85 points of real return before tax. After 20% tax the cash pays 3.0%, a tenth of a point above inflation. After 40% it pays 2.25%, and the buffer loses purchasing power every year. The cash as an asset class guide shows this is unusually good by the standards of the last 92 years, and that the starting real yield decides what a cash sleeve costs.

Two wrappers change the arithmetic. Inside an ISA, interest is untaxed, but the subscription limit is £20,000 a year, so sheltering £150,000 takes 7.5 years of one person's allowance. Inside a pension, interest is also untaxed, but every withdrawal is taxed on the way out: 25% is usually tax-free, capped by the £268,275 lump sum allowance, and the rest is income. At a 20% band a pension withdrawal keeps 85% of its value; at 40% it keeps 70%. The ISA vs pension arithmetic works through why that 25% is the whole edge.

One more rule is about safety. Since 1 December 2025 the FSCS protects £120,000 per person per bank. A £150,000 five-year buffer at one bank is £30,000 over that line. A one-year cash buffer in retirement is nowhere near it.

The order of withdrawal moves the same money through different tax gates

The size question and the wrapper question meet in the order of withdrawal: which pot funds this year's spending, and which pot refills the buffer. The table sets out what each order does mechanically. It's a comparison of tax gates, not a recommendation, and it depends on bands and dates this piece can't know.

The third and fourth rows are where the 2027 change bites. Under the government's policy paper of 26 November 2025, most unused pension funds and death benefits join the estate for Inheritance Tax from 6 April 2027. Until then a pension is the one pot outside the estate, the standard reason for spending everything else first. From April 2027 that reason weakens, and a buffer inside the pension may face Inheritance Tax on top of income tax if it's still there at death. None of this changes the drag arithmetic. It changes which row is cheapest, and for whom.

The wrapper cost also scales with size the way the return drag does. At one year, interest sits under the basic-rate allowance and the FSCS limit is irrelevant. At five years it doesn't, and the tax bill outside an ISA runs to hundreds or thousands of pounds a year on top of 0.88 points of forgone return.

What this evidence can't tell you

Every failure rate here is a backtest, and two of the three datasets are American. Estrada's global results are far less flattering to static portfolios than his US results, and the UK spread of 4.4 points is smaller than the US 6.1, which shrinks the cost of a buffer here. The drag table is arithmetic on long-run averages. It says nothing about any particular decade, and the 0.9% real bill return includes long stretches of financial repression.

The studies test cash against stocks, so a buffer taken out of bonds costs less than the table shows. Tomlinson's point that allocation dominates structure means a retiree comparing a five-year buffer with 80% equities against a 60/40 with none is comparing two allocations, not two buffer sizes.

The behavioural benefit is unmeasured. Evensky's firm has run a cash reserve since 1985 and reports that it helped clients through Black Monday in 1987, the tech crash and the Great Recession. That's testimony, not a controlled comparison. If a visible year of cash is what lets someone hold 60% equities instead of 25%, the drag buys an allocation they'd never otherwise own. If it isn't, it's just drag.

And the tax figures are dated. Every allowance, rate and the 6 April 2027 pension change are as stated by GOV.UK for 2026 to 2027. A Budget can move any of them.

What would change the conclusion

If cash paid a real return near equities, the drag would vanish and buffer size would stop mattering. The spread was 4.4 points in the UK over 115 years. Today's 0.85 points of pre-tax real return on Bank Rate is still 4 to 5 points short of what would make five years cheap.

If stock returns mean-reverted year to year, a rule keyed to last year's return would have something to exploit, and larger buffers would earn their keep in bad sequences. Tomlinson's measured correlation of -0.00069 says they don't. Estrada's third rule, which looks at five-year runs, did best of his three, and five-year mean reversion is a live research question.

If a real behavioural comparison existed, showing retirees with a cash bucket held materially more equities through a crash than retirees without one, the benefit side would finally have a number. Until then, the cost side has three datasets and the benefit side has testimony.

The number to watch is the one neither study prints: how many years of spending your cash represents right now, across every wrapper. LedgerTouch shows the wrapper split; a spreadsheet shows it too. At one year the evidence is relaxed about it. At five it isn't.