Key takeaways
- Over the 92 years to 2025, a fully invested 60/40 portfolio returned 5.31% a year after US inflation. Rebuilt with a 10% cash sleeve, it returned 4.83%.
- That 0.49 point average hides everything interesting. The same 10% sleeve cost 1.01 points a year from 2009 to 2021, and added 0.12 points from 2001 to 2008.
- US Treasury bills beat inflation in only 52 of those 92 years. In 1946 they lost 15.03% of their purchasing power in twelve months.
- In 2022 US equities fell 20.14% in real terms and long bonds fell 20.16%. A 15% cash sleeve cut that year's loss from 20.15% to 17.75%.
- UK Bank Rate paid 4.20% a year in real terms from 1989 to 2008, then minus 1.81% from 2009 to 2021, a 21.1% loss of purchasing power over thirteen years.
A cash sleeve costs a spread, and half of that spread is invisible
Does a permanent slice of cash inside an investment portfolio cost anything, and how much? Over the 92 years from 1934 to 2025, the answer for a 10% sleeve is 0.49 percentage points of real return a year. That figure is also close to useless on its own, and working out why is the whole of this piece.
Here's the test. Take a 60/40 portfolio of US equities and US long-term government bonds, rebalanced once a year, and run it from the start of 1934 to the end of 2025. After US consumer price inflation it returned 5.31% a year. Now rebuild the same portfolio as 54% equities, 36% bonds and 10% three-month Treasury bills. Same years, same annual rebalance. It returned 4.83%.
Cut those 92 years into eras and the gap stops behaving like a constant. From 2009 to 2021 the 10% sleeve cost 1.01 points a year. From 2001 to 2008 it added 0.12 points. The sleeve didn't change. The two things either side of it did.
The mechanism fits on one line. What a cash sleeve costs is its weight multiplied by the gap between what cash earned and what it displaced. Both terms move, they don't move together, and neither one is knowable at the moment you set the weight.
One thing this isn't about: money held for a broken boiler or three months without work. That's a different question with a different answer, and it turns on income volatility rather than returns. The guide to emergency fund size covers it. This piece is about a strategic cash allocation, a sleeve held deliberately inside the invested pot.
What cash as an asset class has actually paid since 1900
Start with the long-run record, because it's the figure most often quoted and the most often misread. The Dimson-Marsh-Staunton database, published each year in the UBS Global Investment Returns Yearbook, runs from 1900 and covers 35 markets. For the United States, the 2026 summary edition puts Treasury bills at "0.5% per year" in real terms over the 126 years to the end of 2025, against 6.6% a year for equities.
The nominal version flatters cash badly. Bills paid 3.4% a year while inflation ran at 2.9%. A dollar of bills became $69 by the end of 2025, and a dollar of equities became $124,854. Over the same period the price level multiplied by 38. Almost all of the headline return on cash was compensation for inflation, and what survived was half a point a year.
That's the standard case against holding cash, and we've made it ourselves in the piece on portfolio cash drag. As an average it's correct. The problem is the averaging. Across the 92 years from 1934 that this piece rebuilds year by year, the real return on bills swung from minus 15.03% in 1946 to plus 7.04% in 1982. One long-run number doesn't describe any decade a real person invested through. Treating cash as an asset class means asking what it paid in the regime you're actually in, which is a harder question than the average admits.
Four cash weights, 92 years, and the shape of the trade
Here's the allocation-versus-outcome table. Each row is the same 60/40 portfolio with a slice moved into three-month Treasury bills, funded proportionally from both equities and bonds, rebalanced annually. Returns are real, after US CPI. The underlying series are Robert Shiller's real total-return indices for US equities and long bonds, and the Federal Reserve Bank of St Louis series for bill rates and consumer prices.
| Cash weight | Equities / bonds | Real return a year | Worst single year | Deepest fall | Real wealth vs no cash |
|---|---|---|---|---|---|
| 0% | 60 / 40 | 5.31% | -24.00% | -36.02% | baseline |
| 5% | 57 / 38 | 5.07% | -22.99% | -34.61% | 19.1% less |
| 10% | 54 / 36 | 4.83% | -21.97% | -33.18% | 34.8% less |
| 15% | 51 / 34 | 4.58% | -20.96% | -31.74% | 47.5% less |
Read the last column before the third. The annual differences look trivial, and over 92 years they compound into something that isn't. A dollar in the fully invested portfolio grew 117.2 times in real terms. The same dollar with a permanent 10% cash sleeve grew 76.4 times, which is 34.8% less wealth at the end.
The risk side is real but modest. Annual volatility fell from 11.64% to 10.61% with the 10% sleeve. The deepest fall shrank from 36.02% to 33.18%. That's the trade in its plainest form: 2.84 percentage points off the deepest fall, for 34.8% of the terminal wealth. Whether that's worth it isn't an arithmetic question, and this piece doesn't pretend to answer it.
The same 10% sleeve cost 1.01 points in one era and paid 0.12 in another
Averaging 92 years is the mistake. Split the record into six stretches and the sleeve behaves completely differently in each. The boundaries below are ours, chosen around two documented turns in policy: the Federal Reserve's wartime peg, and the zero-rate era that followed the financial crisis.
| Era | Real return on bills | Fully invested 60/40 | With 10% cash | Cost of the sleeve |
|---|---|---|---|---|
| 1934 to 1951 | -3.34% | 3.89% | 3.23% | -0.67pp |
| 1952 to 1980 | 0.45% | 3.52% | 3.26% | -0.25pp |
| 1981 to 2000 | 3.08% | 10.24% | 9.56% | -0.67pp |
| 2001 to 2008 | 0.25% | -0.51% | -0.39% | +0.12pp |
| 2009 to 2021 | -1.68% | 8.60% | 7.59% | -1.01pp |
| 2022 to 2025 | 0.26% | 2.35% | 2.21% | -0.14pp |
The chart plots that final column. The first era is the clearest case of cash being an instrument of policy rather than a store of value. In April 1942 the Federal Reserve, at the Treasury's request, pegged short-term bill rates and "implicitly capped the rate on long-term Treasury bonds at 2.5 percent", an arrangement that lasted until the Treasury-Fed Accord of March 1951. Bills paid 0.47% a year nominally over 1934 to 1951 while prices rose 3.95% a year. Holders of cash financed the war at a real loss of 3.34% a year.
The 2009 to 2021 era is the modern echo. The Federal Open Market Committee "decided today to establish a target range for the federal funds rate of 0 to 1/4 percent" on 16 December 2008, and short rates stayed near the floor for most of the thirteen years that followed. Bills returned 0.48% a year nominally against 2.20% inflation, a real loss of 1.68% a year, while the fully invested portfolio compounded at 8.60%. That combination is what produced the worst result in the table.
Then look at 2001 to 2008, the one era where the sleeve helped. Bills earned 0.25% a year in real terms. The 60/40 lost 0.51% a year in real terms across the dot-com unwind and the financial crisis. Cash didn't do anything clever. It simply beat what it displaced.
The starting yield on cash tells you half the answer
If the cost is a spread, an obvious question follows. Cash's yield is the one part of that spread you can see in advance, so does a high starting yield predict a cheap sleeve? Sorting all 92 years by the three-month bill rate in the December before each year gives a partial yes.
| Bill rate at the start of the year | Years | Real return on bills | Fully invested 60/40 | Cost of a 10% sleeve |
|---|---|---|---|---|
| Under 1% | 27 | -2.94% | 3.86% | -0.63pp |
| 1% to 3% | 18 | 0.16% | 8.50% | -0.80pp |
| 3% to 5% | 21 | 1.22% | 4.62% | -0.31pp |
| 5% and above | 26 | 1.90% | 5.23% | -0.27pp |
The starting yield does predict the cash leg cleanly. Years that opened with bills under 1% delivered minus 2.94% a year in real terms. Years that opened above 5% delivered plus 1.90%. That part is close to mechanical, because a low nominal yield leaves nothing to absorb inflation.
It's the second row that breaks the story. Sleeves were most expensive, at 0.80 points a year, when starting yields sat between 1% and 3%. Cash did fine in those years, at 0.16% real. The portfolio it displaced did far better, at 8.50% a year. The cost of the sleeve was high not because cash was bad but because everything else was unusually good, and nobody knew that in the December beforehand.
For scale, the three-month US Treasury bill rate was 3.71% on 25 August 2026, and UK Bank Rate stood at 3.75% on the same day. That puts a reader today in the third row rather than the first.
The strongest case for cash as an asset class is a single year
The serious objection to everything above is that averages measure the wrong thing. What a cash sleeve buys is protection in the specific case where equities and bonds fall together, and that case is rare enough to be invisible in a 92-year mean.
2022 is the cleanest example in the record, and the only year of the 92 in which both equities and long bonds lost more than 20% of their real value. US equities fell 20.14%. US long-term government bonds fell 20.16%. The classic hedge and the thing it was hedging went down by almost exactly the same amount, which is the failure mode set out in the piece on diversification in a crisis. Bills lost 4.15% in real terms that year, badly, and still lost a fifth as much as everything else.
The arithmetic of that year: a fully invested 60/40 lost 20.15% in real terms, a 10% cash sleeve brought it to 18.55%, and a 15% sleeve brought it to 17.75%. So the counter-case is worth 2.4 percentage points in the worst joint fall in the sample. Whether 2.4 points in one year justifies 34.8% of terminal wealth over 92 depends entirely on what a person needs the money for and when.
There's a weaker version of the objection worth naming too. Cash beat the fully invested 60/40 in 23 of the 92 years, so it isn't as though the sleeve is dead weight in every period. It's dead weight in most of them.
What a UK saver earned on cash across the same eras
The long series above are American, because that's where a century of clean data exists. UK cash tells the same regime story over a shorter window. Compounding the official Bank Rate daily and deflating by the ONS consumer prices index gives a real return of 4.20% a year across 1989 to 2008, and minus 1.81% a year across 2009 to 2021.
Thirteen years at minus 1.81% is a 21.1% cut in purchasing power. 2022 alone took another 8.18% in real terms. Since 1989 the whole-period figure is 4.32% nominal and 1.48% real, and that average is doing the same concealing work as the American one.
Two things make this generous rather than pessimistic. Bank Rate is a policy rate, not a deposit rate. In July 2026 the Bank's own series for sterling instant access deposits from households put the average at 2.11%, against a Bank Rate of 3.75%, a gap of 1.64 points. Tax comes off the interest on top of that. A saver's real return sat below every figure in this section.
What this test can't tell you
This is a backtest of one country's history, not a forecast, and several things about it deserve to be said plainly.
The sample is 92 years of one country. The DMS database covers 21 countries with continuous investment histories, and equities were the best-performing asset class in all of them, so the direction of the result probably travels. The size of it is another matter. A cash sleeve costs less in a market where the alternative earns less, and this test only ran on the American one.
The era boundaries are ours. Two of them sit on documented policy events, and the rest are judgement. Move them and the numbers move. That's a real weakness of any regime study, and the reason the sorted-by-starting-yield table is in the piece as well.
The model rebalances annually with no trading costs, no spreads and no tax. All of those work against the version with more moving parts, so the fully invested portfolios in the table are flattered a little. The Treasury bill returns used here are wholesale rates, and the UK deposit gap above is the reminder that a retail cash line earns less. And a single number for volatility hides the thing people actually experience, which is covered in the piece on volatility versus drawdown.
Nothing here measures behaviour. If a cash sleeve is what keeps someone from selling equities in the twelfth month of a 36.02% fall, its value doesn't appear in any of these columns. That's not a small omission. It's just not something a return series can see.
What would change the conclusion
Three things would move it, and they're worth watching in that order.
The first is the real return on cash. Every era in the table where the sleeve was cheapest is one where bills beat inflation, though 1981 to 2000 shows that on its own isn't enough. Bills have beaten inflation in only 52 of 92 years, so a positive real yield isn't the default state, and if it reverts the sleeve gets expensive again quickly.
The second is the correlation between equities and bonds. The case for cash was strongest in 2022 precisely because bonds stopped working. If stocks and bonds go back to falling together in inflation shocks, the sleeve's job stops being drag and starts being the only diversification left. If they don't, bonds do that job at a higher expected return.
The third is the one nobody can observe. The cost of a sleeve is the spread against what it displaced, and that half only becomes known afterwards. A tracker like LedgerTouch can measure the drift between the sleeve you intended and the one you actually hold, which is the observable part. The other part isn't. Anyone setting a 5% or a 15% weight today is pricing a number they cannot see, and that's a reason to hold the estimate loosely rather than a reason to pick a side.