Across seven US stock market crashes from 1929 to 2022, T-bills gained in every window, while the other havens had exceptions: 10-year Treasuries beat T-bills in five crashes but lost 15.1% in 2022, gold rose in five, and trend following gained in four of the five since 1987, including 34.9% in 2022 (our calculation, data to September 2026). Our crash study measured how far each asset class fell; this sequel asks which ones paid while shares fell, and which only fell less.
We measure every haven from the month-end peak of US stocks to their month-end trough, in dollars since 1929 and in euros since 2000. Then we measure what each haven cost in the years without a crash, because that cost compounds too.
Key takeaways
- Cash provides liquidity even without a price gain. T-bills gained in all seven US crash windows and can fund spending or rebalancing; those are separate uses from a hedge that appreciates when equities fall.
- The type of crash changes the result. In 2022, 10-year Treasuries lost 15.1% and euro government bonds 16.7%, while the trend factor gained 34.9% in dollars before fees; test inflation shocks as well as recessions.
- Judge protection inside the whole portfolio. Every 80/20 haven mix reduced the worst fall in our scenarios, but every mix also reduced growth over 2010-2026; compare that trade-off with your spending needs and implementation costs.
Rose, held or fell less: what is the difference?
An asset "rose" when it beat one-month T-bills over the crash, "held" when it ended between zero and the T-bill return, and "fell less" when it lost money but less than shares. The difference matters for growth. A holding that rises during a crash can offset losses elsewhere and provide additional money for rebalancing. Cash or an asset that falls less can also fund purchases, though without the same gain; compare the amount available, its intended use and the costs of holding it.
We measured each crash in the table of US bear markets in our crash study, from the month-end peak to the month-end trough of the US stock market's total return. These are the same windows as in our study of Nassim Taleb's barbell, plus the 1987 crash. Month-end values miss the deepest days: the crash study dates the S&P 500's 1987 fall at 33.5% from its August peak to its December trough, against 29.9% for US stocks between the August and November 1987 month ends.
Which assets rose in each US crash since 1929?
Treasuries, gold and trend following each rose in most crashes we measured, but each failed at least once, and T-bills never lost money.
| Crash (month-end peak to trough) | US stocks | 10-year Treasuries | 1-month T-bills | Gold (USD) | Trend following |
|---|
| 1929-32 (Aug 1929 - Jun 1932) | -83.7% | +9.2% rose | +6.0% held | 0.0% held | n/a |
| 1973-74 (Dec 1972 - Sep 1974) | -46.5% | +0.4% held | +13.5% held | +137.5% rose | n/a |
| 1987 (Aug - Nov 1987) | -29.9% | +1.6% rose | +1.4% held | +1.5% rose | -1.3% fell less |
| 2000-02 (Aug 2000 - Sep 2002) | -45.0% | +29.5% rose | +7.3% held | +16.9% rose | +61.8% rose |
| 2007-09 (Oct 2007 - Feb 2009) | -50.3% | +20.1% rose | +2.2% held | +20.6% rose | +32.5% rose |
| 2020 (Jan - Mar 2020) | -20.2% | +8.8% rose | +0.2% held | +1.6% rose | +10.8% rose |
| 2022 (Dec 2021 - Sep 2022) | -24.8% | -15.1% fell less | +0.6% held | -8.2% fell less | +34.9% rose |
Returns in US dollars over each window, before costs and taxes. US stocks: CRSP value-weighted total return. Trend following: AQR's time series momentum factor plus T-bill interest, available from 1985.
Three exceptions shape the record. In 1929-32, gold could not rise: its official price was $20.67 an ounce until the spring of 1933, according to Federal Reserve History. Prices fell 21.4% over that window, so T-bills and Treasuries gained even more in real terms. In 1973-74, Treasuries returned 0.4% while US consumer prices rose 19.1%, and gold's 137.5% gain was out of reach for Americans, who could not legally own gold until 1974.
10-year Treasuries: +1.6%, +29.5%, +20.1%, +8.8%, -15.1%. Gold: +1.5%, +16.9%, +20.6%, +1.6%, -8.2%. Trend following: -1.3%, +61.8%, +32.5%, +10.8%, +34.9% (1987, 2000-02, 2007-09, 2020, 2022). Source: our calculations from the Kenneth French data library, Robert Shiller's data (10-year Treasury total return), the World Bank Pink Sheet (gold, 1987), gold in euros from our crash study converted at ECB month-end rates (gold from 2000), and AQR's Time Series Momentum factors; chart data (CSV); monthly inputs (CSV).
Why did bonds fail in 2022?
Bonds fell with shares in 2022 because the fall came from rising inflation and interest rates, which lower bond prices, rather than from a recession that pulls yields down. US consumer prices rose 6.5% from December 2021 to September 2022, and the 10-year Treasury yield rose from 1.47% to 3.52%, according to Shiller's data.
In the crashes where Treasuries paid most, the yield fell: from 5.83% to 3.87% in 2000-02, from 4.53% to 2.87% in 2007-09 and from 1.76% to 0.87% in 2020. It also rose in 1973-74, from 6.36% to 8.04%, when Treasuries returned 0.4% against 13.5% for T-bills. Our article on why bonds protected in 2008 but failed in 2022 traces the mechanism over longer windows.
December 2021 to September 2022: US stocks -24.8% in dollars and -11.5% in euros; 10-year Treasuries -15.1%; euro government bonds -16.7%; EUR-hedged global government bonds -14.1%; T-bills +0.6%; euro cash -0.2%; gold -8.2% in dollars and +6.7% in euros; the dollar +16.2% against the euro; trend following +34.9%. Source: our calculations from the same sources, plus the iBoxx € Sovereigns, FTSE WGBI (EUR-hedged, corrected series), Solactive Euro Overnight Return and S&P 500 total return levels in euros used in our crash study; chart data (CSV).
For a euro investor, euro government bonds lost more than US shares over those nine months. The dollar's rise cushioned the S&P 500 in euros and turned gold's dollar loss into a euro gain. These windows differ from the crash study's three-year windows, so its 21.2% fall for euro government bonds (December 2020 to December 2022) and the 16.7% here measure different periods.
What did a European investor see in the crashes since 2000?
In euros, gold and the US dollar rose in the 2007-09, 2020 and 2022 crashes, euro government bonds rose in 2000-02 and 2007-09 but fell in 2020 and 2022, and euro cash held or slipped slightly.
| Crash (US month-end peak to trough) | S&P 500 in euros | Euro government bonds | Global government bonds, EUR-hedged | Euro cash | Gold in euros | US dollar vs euro | Trend following, euro cash |
|---|
| 2000-02 | -50.1% | +16.7% | +19.4% | +9.0% | +5.6% | -9.7% | +64.2% |
| 2007-09 | -44.0% | +9.2% | +10.1% | +5.0% | +37.8% | +14.3% | +36.0% |
| 2020 | -18.9% | -2.1% | +1.4% | -0.1% | +2.5% | +0.9% | +10.4% |
| 2022 | -11.5% | -16.7% | -14.1% | -0.2% | +6.7% | +16.2% | +33.9% |
Returns in euros over the same US windows, before costs and taxes. Euro cash: EONIA compounded deposits to 2005, Solactive Euro Overnight Return from 2006. Trend following, euro cash: AQR's excess return plus euro cash, our construction for a euro-hedged version. Data: CSV.
The dollar is a two-sided haven for a euro investor. It rose 14.3% against the euro in 2007-09 and 16.2% in 2022, but fell 9.7% in 2000-02, when it deepened the S&P 500's loss in euros to 50.1%. Holding US assets unhedged therefore changes what a crash does to a euro portfolio, in either direction.
Does trend following pay in every crash?
No: it gained in four of the five crashes since 1987, but lost 1.3% in 1987, when shares fell too fast for a 12-month trend signal to turn. AQR describes its factor as "a 12-month time series momentum strategy with a 1-month holding period" across "58 underlying liquid instruments" in equity indices, currencies, commodities and government bond futures (AQR). A trend follower goes short when markets have been falling, so it profits from long declines such as 2000-02, 2007-09 and 2022, and less from sudden ones.
Two limits apply. The factor is a research series before fees, trading costs and taxes, not a fund. Its 1985-2009 returns are the sample of the 2012 paper that documented the strategy, so they flatter it: from January 2010 to May 2026 the factor, with T-bill interest, compounded at 5.94% a year against 14.39% for US stocks. Our study of Cliff Asness and AQR covers AQR researchers' comparison of trend following with put options as crash protection.
What does a haven cost in normal years?
Every haven earned less than shares outside the crashes, and that gap compounds: from 1985 to May 2026, US stocks grew 12.0% a year, against 6.05% for 10-year Treasuries, 6.64% for gold and 3.21% for T-bills.
| Asset | Growth a year, 1985-2026 | Growth a year outside crash windows | Growth a year, 2010-2026 |
|---|
| US stocks | 12.00% | 20.46% | 14.39% |
| 10-year Treasuries | 6.05% | 5.74% | 2.15% |
| 1-month T-bills | 3.21% | 3.29% | 1.38% |
| Gold (USD) | 6.64% | 6.49% | 8.88% |
| Trend following | 15.65% | 14.14% | 5.94% |
Our calculations, in US dollars, before costs and taxes; 1985-2026 runs from January 1985 to May 2026, 497 months of which 55 fall inside the five crash windows. The crash windows are defined by the stock market's own peaks and troughs, so the months outside them include every rebound and overstate shares' normal-year return. From July 1926 to September 2026, US stocks grew 10.36% a year, Treasuries 4.70% and T-bills 3.29%. Data: CSV.
The fair test is the whole portfolio. Mark Spitznagel, founder of Universa and the subject of our tail-hedging study, states it in Safe Haven (2021, p. 20): "If a strategy cost-effectively mitigates a portfolio's risk, then adding that strategy raises the portfolio's CAGR over time." We applied that test to five scenario mixes, each rebalanced monthly.
Growth a year 1985-2026 / 2010-2026 / worst fall 1985-2026: 100% US stocks 12.00% / 14.39% / -50.3%; with 20% Treasuries 11.04% / 12.08% / -40.2%; with 20% T-bills 10.40% / 11.87% / -42.2%; with 20% gold 11.29% / 13.61% / -39.3%; with 20% trend following 13.11% / 13.06% / -38.6%. Source: our calculations from the Kenneth French, Shiller, World Bank and AQR data above; 1985-2026 scenarios (CSV); 2010-2026 scenarios (CSV). These are scenarios, not recommended weights: one 20% share, monthly rebalancing and no costs, fees or taxes.
By Spitznagel's test, no haven passed over 2010-2026 at this weight: every mix grew more slowly than stocks alone. Over 1985-2026 only the trend mix grew faster, turning $1 into $164 against $109 for stocks, and that period includes the sample in which the strategy was found. The T-bill mix shows the cost of a pure cushion: it cut the worst fall to 42.2% and gave up 1.6 points of growth a year.
What does copio take from these crashes?
A useful defence needs a defined job: fund spending, offset market losses or preserve long-run growth. Cash and T-bills can fund purchases after a fall without first rising in price. Bonds, gold and trend can add gains during some crashes, but none beat cash in every observed crash window.
In our 2010-2026 80/20 scenarios every haven reduced both the worst drawdown and compounded growth. That is a historical trade-off, not proof that the assets failed to diversify or that a spending reserve has no purpose. Copio's growth objective requires testing the whole allocation, including costs and the withdrawals it must support.
Three checks for your portfolio
- Assign a job to each defensive holding. Write whether it is there to fund spending, provide rebalancing cash, offset a market loss or diversify long-run returns. Compare accessible cash with planned withdrawals before treating it as available to buy shares after a crash. One reserve cannot be spent twice.
- Test different shocks in your spending currency. Compare each holding and the combined portfolio in 2020 and 2022, using the same dates and including currency hedges. Record whether each holding rose, stayed near its starting value or fell, and inspect bond duration and credit exposure. The dollar helped a euro investor in 2022 and hurt in 2000-02; the currency printed in a fund's name does not establish that its exposure is hedged.
- Set a whole-portfolio hurdle before adding protection. Choose the loss or spending shortfall you want to reduce, then compare growth, drawdown and liquidity with and without the holding over the same periods. Include fees, taxes and renewal or trading costs where relevant, and test the weaker periods as well as the headline crash. The trend research factor is not a purchasable fund; obtain an actual product's rules and costs before using its historical figures in a decision.
The portfolio checker can help explore supported funds. Our put-option guide explains strike, expiry and premium constraints; our gold study includes its long losing stretches.
FAQ
Which assets went up in stock market crashes?
Across seven US crash windows from 1929 to 2022, one-month T-bills gained in every window in dollars. Ten-year Treasuries beat cash in five, gold rose in five, and the trend factor gained in four of its five available windows. No other studied haven beat cash in every observed window. In 2022 Treasuries lost 15.1% and gold 8.2%, while trend gained 34.9% before fees.
Why did bonds not protect in 2022?
Because the fall in shares came with rising inflation and interest rates. US consumer prices rose 6.5% from December 2021 to September 2022, and higher yields lowered bond prices: 10-year Treasuries lost 15.1% and euro government bonds 16.7% over those months, more than the 11.5% that the S&P 500 lost in euros.
Is gold a safe haven in a crash?
Often, not always. Gold in dollars rose in five of the seven crashes we measured, from 1.5% in 1987 to 137.5% in 1973-74, but lost 8.2% in 2022. In 1929-32 its price was fixed by law at $20.67 an ounce, and US residents could not legally own gold during 1973-74. In euros, gold gained in all four crashes since 2000, though in 2000-02 by less than euro cash.
Does trend following protect against crashes?
In four of the five crashes since 1987 AQR's time series momentum factor gained, by 10.8% to 61.8% with T-bill interest added and before fees. It lost 1.3% in the fast 1987 crash, when there was no downtrend to follow yet. It is a research factor, not a fund, and its 1985-2009 returns are the sample in which the strategy was first documented.
What does holding a safe haven cost?
Compare opportunity cost, fees, taxes and implementation costs with the job it performs. In our 1985-May 2026 scenarios, moving 20% from US stocks to Treasuries, T-bills or gold lowered annual growth by 0.7 to 1.6 points and reduced the worst fall from 50.3% to 39.3-42.2%, before costs and taxes. A spending reserve can still be useful even when it lowers historical portfolio growth.
