Moving to Cash in a Crash: 10 Bear Markets Tested

Key takeaways

• Across 10 US bear markets since 1926, moving to cash at the month the market was 20% below its high beat staying invested at the eventual trough in 8 of them, by a median of 15.7%.
• One year past that same trigger cash was ahead in 4 of 10 episodes. Ten years past it, in 1 of 8, and the median gap was 36.5% behind.
• Selling at 20% down and buying back only when the index regained its old high lost in 10 episodes out of 10, by between 5.9% and 23.1%.
• Over those out-of-market windows equities gained 25.8% to 37.3%. Cash returned 0.0% to 20.9%, and closed the gap in none of the 10.
• A UK instant access account compounded 9.07% between June 2022 and July 2026 while UK CPI rose 17.32%, a real loss of 7.04% over 49 months.

Cash is king for a few months, and then the bill arrives

You've watched a quarter of your portfolio go, someone has told you that cash is king in a crisis, and you want to know whether moving to cash in a crash has actually worked.

On a century of US data it's true for a few months and then it stops being true. Take the moment the market closes a month 20% below its previous high, which is roughly when the headline gets written, and move the whole equity holding into one-month Treasury bills. Measured at the eventual bottom, that trade was ahead in 8 of the 10 bear markets since 1926, by a median of 15.7%. Measured a year on, it was ahead in 4 of 10. Measured ten years on, in 1 of 8, and the median outcome was 36.5% behind.

The bottom is the only date that flatters the trade. It's also the one date nobody has in advance. Once you replace it with a rule you could actually have followed, the record gets worse than the medians suggest, and it gets worse in a way that doesn't depend on how the next decade goes.

The test: 10 bear markets, one exit rule, no forecasting allowed

The data is Kenneth French's monthly series, July 1926 to June 2026, which is 1,200 months. The equity leg is his market factor: the "value-weight return of all CRSP firms incorporated in the US and listed on the NYSE, AMEX, or NASDAQ", dividends included. The cash leg is the one-month Treasury bill rate that French subtracts from it to get the market's excess return. Using the same file for both legs means the comparison isn't smuggling in a difference between two vendors' conventions.

A bear market here is a month-end fall of 20% or more from the running peak of that total-return index. The trigger is the first month-end at or below minus 20%. The episode closes when the index regains its old peak. That gives 10 episodes in 100 years, starting in October 1929 and ending with the one that began in June 2022.

Two of the 10 are awkward, and it's worth saying so early. In June 1962 and in March 2020 the trigger month was also the trough month. A monthly series can't see the week inside March 2020 when a real person would have sold, so those two episodes start the cash trade at a disadvantage the data can't resolve.

Moving to cash in a crash was right at the trough in 8 of 10 bear markets

Judged only against the bottom, the trade looks strong. After the October 1929 trigger the market fell another 78.6% before it stopped, on its way to a peak-to-trough loss of 83.7%. After the September 2008 trigger it fell another 35.7%, and cash was 55.7% ahead at the February 2009 low. Those are not marginal wins. If you had sold at minus 20% and bought back at the exact bottom, you'd have done very well twice, and reasonably well six more times.

The chart below tracks what happens to that same trade as the measuring date moves out. At the trough the median advantage is 15.7%. At one year it's already minus 4.7%. At five years it's minus 31.1% across the 9 episodes old enough to measure, and at ten years minus 36.5% across 8.

Nothing in that decay is surprising once you write it down. Cash pays a low, positive number every month. Equities pay a higher number on average and a violently negative one occasionally. Sell after the violent part and you've swapped the higher average for the lower one, right at the point where the higher one has already been paid for.

Waiting for the all-clear lost in 10 episodes out of 10

The trough is unknowable, so here's a rule that isn't. Sell when the market is 20% down. Buy back when the index makes a new high, which is the closest thing to an all-clear that markets issue. It's the rule people describe when they say they're waiting for things to settle, and it never requires a forecast.

It lost every time.

The median shortfall is 19.3%. The best case, November 1973, still lost 5.9%, and it's the best case because cash returned 20.9% over the 37 months out, more than over any other absence in the table. The worst, March 2020, lost 23.1% in four months out of the market. Length of absence turns out to be a poor guide to damage: 183 months out from 1929 cost 19.2%, and four months out from March 2020 cost more.

The cost of waiting is fixed the day you come back, not the day you sell

That last line looks strange until you write out the arithmetic, and the arithmetic is the whole piece.

If you leave the market for a window and then return, your result relative to never selling is the cash return over that window against the equity return over the same window. Whatever happens afterwards happens to both versions of you, identically. So the horizon you measure at is irrelevant. The cost of waiting is settled at the moment you get back into the market, and nothing after that changes it.

Now look at what the all-clear rule does to that window. Re-entering at the old high means, by definition, that equities have climbed back everything they lost while you were out. Leaving at 20% down means they had to rise about 25% to get there. That's not a forecast, it's the shape of a percentage: a fall of a fifth needs a rise of a quarter to undo it. Across the 10 episodes equities gained between 25.8% and 37.3% over the absence. Cash returned between 0.0% and 20.9%. It never made up the difference, and the reason it never did is that the rule sets the hurdle before the waiting starts.

This is the same asymmetry that governs bear market recovery time, viewed from the other end. The recovery a stayer waits through passively is the recovery a seller has to beat with a savings rate.

1929 is the case for the defence, and it's a real one

One episode says cash was king, and it says it loudly. Move to cash in October 1929 and stay there, and you were 217.8% ahead of the market at three years and still 21.8% ahead at ten. In real terms cash was up 31.5% over that decade while equities were up 8.0%. That isn't a rounding error or a quirk of month-end sampling. It's ten years of being right, in the worst equity market the US has recorded.

Two things sit alongside it. At twenty years from the same trigger cash was 44.7% behind, so the advantage was a long episode rather than a permanent state. And the all-clear version of the trade still lost 19.2%, because the index didn't regain its August 1929 peak until January 1945. It came within 0.48% of it in December 1944 and stalled. An investor holding out for a clean new high in the 1930s was waiting 183 months for a signal that kept not arriving.

The 2008 episode runs the other way and is worth putting next to it. Cash led by 55.7% at the February 2009 bottom, which is about as good as this trade gets outside the Depression. Over the ten years that followed the September 2008 trigger, cash returned 2.7% in total while US consumer prices rose 15.4%. That's an 11.0% real loss, against 221.2% nominal for the equities that were sold. Being right about the crash and wrong about the decade is the characteristic failure of moving to cash in a crash, and it isn't rare.

What UK cash actually pays, measured against UK prices

The US series above uses Treasury bills, which is not what a UK saver holds. The UK numbers are worse, for a reason that has nothing to do with markets.

Bank Rate was 3.75% in July 2026. The average rate on sterling instant access deposits from households, the Bank of England's IUMB6VJ series, was 2.11% in the same month. That gap of 1.64 points is the part of the story the phrase leaves out: the cash you hold is not the cash the central bank pays. From June 2022, the trigger month of the most recent episode, to July 2026, that instant access series compounded to 9.07%. UK CPI rose 17.32% over the same 49 months. In purchasing power that's a loss of 7.04%.

Stretch it back to January 2011, where the series begins, and instant access cash returned 16.33% against 56.52% for CPI. That's 25.68% of purchasing power gone over 15 and a half years, or 1.90% a year. None of that period contains a crisis exit. It's what the safe option costs when nothing happens at all, which is the argument made at length in our piece on portfolio cash drag.

The FCA's Financial Lives survey, fieldwork February to June 2024 across 17,950 interviews, found that 61% of UK adults with £10,000 or more in investible assets held all or at least three-quarters of it in cash savings. That was 55% in 2020 and 58% in 2022, so the share has risen. Among adults with £10,000 or more in cash and no investments who hadn't taken advice, 59% agreed that money in cash savings decreases in value because interest rates usually don't keep pace with inflation, down from 67% in 2022.

What this test can't tell you

The sample is 10 episodes in one country over 100 years, and 10 is a small number for a claim this general. Two of the 10 are still running out their ten-year windows, which is why the medians are taken across 8 and 9 episodes at the longer horizons rather than all 10.

Monthly data hides the part of a crash that people actually experience. March 2020 shows as a 20.2% month-end drawdown with the trigger and the trough in the same month. The intramonth path was worse, and a real seller would have transacted somewhere inside it at a price this test can't see.

There is no tax and no dealing cost in any of these figures. Selling a taxable equity holding after a 20% fall crystallises whatever gain or loss is there, and that can dominate the arithmetic for a particular person. The exit is also all-or-nothing at exactly minus 20%. A shallower trigger fires more often and a deeper one fires later, and both change the numbers. The critics of this kind of backtest are right about the general point: a rule tested on the only history we have is a rule fitted to it, and the next episode is not obliged to resemble the last 10.

What the test does not depend on is a forecast, which is the one thing in its favour. The all-clear result holds whatever equities do next, because the shortfall is complete before the measurement starts.

What would change the conclusion

A cash rate that beats equities across a multi-year absence would make moving to cash in a crash pay, and it has happened once in 100 years. October 1929 is that case, and anyone arguing the claim should argue it from there rather than from 2008 or 2022, both of which look supportive for a few months and then don't.

The mechanism gives you the threshold to watch, because it's arithmetic rather than prediction. An exit at 20% down sets a recovery of roughly 25% as the hurdle, and cash has to clear that hurdle over however long the absence runs. At the 3.75% Bank Rate of July 2026 that takes something over six years of compounding, which is longer than 9 of the 10 recoveries in the table took. A materially higher cash rate, or a materially slower recovery than any in this sample, would be the combination that changes the answer.

Two things narrow the scope rather than change it. If you're drawing an income from the portfolio the arithmetic is different, because selling in a drawdown is forced rather than chosen. And if the question is what to do with money you already hold in cash rather than whether to create some, that's a separate calculation, one we've run against buying the dip and against rebalancing in a crash. LedgerTouch tracks drawdown from peak per holding, which is the number this trigger is built on.