Managed Futures in a Crisis: Six Drawdowns Tested

Key takeaways

• Across the six developed-equity falls of 20% or more since July 1990, AQR's time series momentum factor gained in every one, from 10.3% in the 2020 crash to 53.9% through 2000-02.
• Its monthly correlation with world equity was -0.175 across 431 months, but -0.491 inside the 70 crisis months and +0.005 across the other 361.
• The equity leg lost money in 3 of the 6 crises. The fixed income leg was positive in all 6, including the 2022 window, when bonds themselves were falling.
• A 20% sleeve would have cut the 2007-09 loss on a world equity portfolio from 50.24% to 38.71%, and the full-sample maximum drawdown from 53.62% to 41.85%.
• The same factor earned 19.59% a year in the 1990s and 3.40% a year since 2020, and in November 2021 it sat 27.91% below its February 2016 high.

Managed futures made money in all six equity crashes since 1990

You've been told managed futures earn their keep when shares fall. You want to know whether the record actually says that. It does, six times out of six. What the record doesn't say is that the help was reliable in size, or that it was free.

The test runs across every fall of 20% or more in developed-world equity since July 1990. There have been six. Over the months each fall occupied, a standard trend following rule returned 23.16%, 10.26%, 53.92%, 26.76%, 10.50% and 22.62%.

Those returns come from AQR's published time series momentum dataset, the monthly series behind the 2012 paper by Moskowitz, Ooi and Pedersen. The rule is plain. Hold each of 58 futures and forward contracts long if its own past 12-month excess return was positive, short if it was negative, then rebalance the following month. No discretion, no view.

So the crisis alpha claim survives contact with the data. The interesting part sits underneath it. The protection was uneven, it came from a different place than the pitch implies, and between crises it cost real money.

The six windows come from a rule set in advance, not from memory

Choose your crisis dates by hand and you'll choose the ones that flatter the answer. These come from a rule fixed before anything was measured.

The reference series is the Fama/French Developed market portfolio: every listed share in that universe, value-weighted, dividends included, in US dollars, daily from 2 July 1990. A running maximum is carried forward from the first day. A window opens at the running maximum preceding any fall of 20% or more from it, and closes at the lowest close before that maximum is regained. Nothing in that rule asks what counted as a crisis. It produces six windows, the same six used in our piece on diversification in a crisis.

One convention matters here. The managed futures series is monthly, so each equity window is rounded out to whole months, from the month end before the peak to the month end after the trough. That's a slightly wider window than the equity fall itself. It's stated so you can discount it, and it cuts both ways.

The chart above plots that last column. Six bars, all of them positive, and a spread between the best and the worst of more than 43 percentage points.

The correlation turned negative exactly when it needed to

An asset that's uncorrelated on average is worth little if the correlation snaps upward in the fall. That's the failure this site has documented under correlation instability, and it's the reason a long-run correlation table flatters almost every diversifier ever sold.

Managed futures did the opposite. Over the 431 months from July 1990 to May 2026, the factor's monthly correlation with world equity was -0.175. Inside the 70 months that fall within the six crisis windows it was -0.491. Across the other 361 months it was +0.005, which is to say nothing at all.

Read down the middle column. Every leg moved against equity harder in a crisis than it did on average, and the diversified factor moved hardest of all. That last point is worth sitting with. The whole is more negatively correlated than any of its parts, which follows from the legs being only loosely tied to each other: their pairwise correlations run from +0.081 to +0.288 across the same 431 months.

The monthly numbers say the same thing in a flatter way. The factor's average month across the whole sample was +0.75% excluding crisis months and +1.94% inside them, against -2.49% a month for world equity in those same crisis months.

The crisis alpha came from bonds and commodities, not from being short shares

Here's where the folklore breaks. Most people picture a trend follower in a crash as a fund that's short the stock market, cashing in on the fall. That isn't what paid.

AQR publishes the same factor split into its four legs. Run each leg through the same six windows and the attribution is not the one on the tin.

The equity leg lost money in three of the six. It fell 20.26% through 1998, 46.22% through the 2020 crash and 30.06% through 2021-22. Each leg is a high-volatility construction in its own right, which is why the percentages look so violent: the equity leg ran at 26.82% annualised volatility over the sample and the fixed income leg at 28.67%, against 12.51% for the four combined. The sign is the point, not the size. In half the sample the part of the strategy sold as the crash hedge was the part losing money.

The fixed income leg was positive in all six, and by the widest margins: 87.17% through 2000-02 and 85.57% through 2021-22. The second of those is the one to notice. Bonds weren't a refuge across that stretch: the Bloomberg US Aggregate Bond Total Return Index averaged -0.36% a year over the five years to 31 December 2025, a span that contains this window. A leg that simply held bonds could not have returned 85.57% inside it, so the signal had already flipped short.

That's the mechanism the name hides. Trend following doesn't hedge equities. It follows whatever has been moving, in either direction. The commodity leg was positive in all six windows too, between 7.59% and 28.97%. The currency leg was positive in five.

AQR's century-long study reaches the same place from a different direction, and explains it. Simulating the strategy back to 1880, it reports positive returns in "8 out of 10" of the largest drawdowns for a 60/40 portfolio. The reason its authors give is about timing rather than about equities: "most bear markets have historically occurred gradually over several months, rather than abruptly over a few days, giving trend followers an opportunity to position themselves short after the initial market decline and profit from continued market declines". The average peak-to-trough length of those 10 drawdowns was roughly 15 months. The two windows where the trend bar sits below zero are the 1937 recession and the 1987 crash, and on the second the authors are explicit: in bear markets that occur very rapidly the strategy "may not be able to take positions quickly enough to benefit from sharp market movements in those environments".

What a 20% managed futures sleeve did to the drawdowns

Correlations are abstract. Portfolio arithmetic isn't.

Take a portfolio of world equity, move 10% or 20% of it into the factor, rebalance monthly, and let the sleeve earn its excess return plus the one-month bill rate from the same dataset. Through the 2007-09 window the loss goes from 50.24% at no sleeve, to 44.71% at 10%, to 38.71% at 20%. Measured over whole months rather than exact days the equity fall is 50.24% rather than the 57.03% it reached peak to trough, so the rounding hands back part of the fall before the sleeve does anything.

Across the full 431 months, all-equity returned 8.58% a year with 15.05% volatility and a 53.62% maximum drawdown. With a 20% sleeve that becomes 10.05% a year, 11.86% volatility and a 41.85% maximum drawdown. Higher return, lower risk, from a sleeve that was flat or losing through most of the period. The same shape shows up when you test any uncorrelated diversification benefit against a balanced portfolio.

AQR's independent version, on US data from January 1880 to December 2016, with a 2% management fee and a 20% performance fee taken off the trend sleeve, puts an 80/20 blend of a 60/40 portfolio and the trend strategy at 4.8% a year in excess of cash against 4.1% for the 60/40 alone, with a maximum drawdown of 50.2% against 62.3% and a net-of-fee Sharpe ratio of 0.55 against 0.39. Two samples, two constructions, the same direction of travel.

The counter-case: the factor is gross of fees, and the funds are not

Now the objection, and it's a serious one. Everything above is a research factor. Nobody owns it. Critics of managed futures have argued for years that the gap between the factor and the investor is where the entire return goes.

Bhardwaj, Gorton and Rouwenhorst made that case in a 2008 National Bureau of Economic Research working paper. Over 1994 to 2007, they find, "CTA excess returns to investors (i.e., net of fees) averaged 85 basis points per annum over US T-bills, which is insignificantly different from zero". Gross of fees they put the figure at 5.4%. On their estimate the managers captured most of what the strategy produced.

There's a reason this piece tests a factor rather than an industry benchmark. Societe Generale publishes an SG Trend Index built from the 10 largest trend-following managers open to new investment, equally weighted, reconstituted every 1 January and calculated net of all fees. Being net of fees, it's closer to what an investor actually received. It also starts on 1 January 2000, so it can't reach the first two windows here.

Fees are still the largest visible line. AQR's Managed Futures Strategy Fund, whose two longest-serving portfolio managers have run it since 5 January 2010, is a US mutual fund built on the same idea. Its summary prospectus dated 1 May 2026 reports total annual operating expenses of 3.09% a year on the Class I shares, or 1.25% once dividends on short sales and interest expense are stripped out. Portfolio turnover in the most recent fiscal year was 412% of the average value of the portfolio.

And yet the fund record doesn't settle it the way the critics expect. Over the ten years to 31 December 2025 that fund returned 3.58% a year before taxes, against 2.18% a year for the ICE BofA US 3-Month Treasury Bill Index. That's 1.40 points a year over cash, after every fee in that prospectus. The AQR factor over those same ten calendar years returned -0.45% a year in excess of cash. The live fund beat the textbook.

Which means the tidy version of the objection, that the gross factor flatters the strategy, doesn't hold on this evidence. What both measures agree on is narrower and harder to dismiss: the last ten years were poor.

Between crises, the strategy has been expensive to hold

The decade table is the part the marketing leaves out.

The volatility barely moves, which is by design, because positions are scaled to a risk target rather than to a capital amount. The return moves a great deal. Over the whole 431 months the factor made 11.12% a year in excess of cash at 12.51% volatility, and it was positive in 257 of those months, or 59.6% of them. Inside crisis months it was positive in 48 of 70.

The path was worse than any of those averages suggests. From its high in February 2016 the factor fell, and by November 2021 it still sat 27.91% below that high. That's nearly six years of an insurance policy quietly bleeding, on a promise about a crisis that hadn't arrived.

The year 2020 makes the point in miniature. The factor gained 10.50% over the two crisis months and finished the calendar year down 6.74%. The crash paid; the year didn't. Trend following across futures is a different animal from the momentum premium in single shares, which sorts stocks against each other rather than each market against its own past, but both share that habit of giving back in calm what they earned in dislocation.

What this evidence can't tell you

Six crises is a sample of six. That is not a lot to hang a strategy on, and no statistical test on six observations is going to be persuasive. The correlation figures rest on more, 431 months, but the crisis correlations rest on 70.

The factor is a rule applied to historical prices, not a fund you could have bought. Nothing in the dataset's documentation deducts fees or trading costs, and the live fund built on the same idea turned over 412% of its portfolio in a year. AQR also notes that the data sources and methodology behind the published series may differ from the original 2012 paper.

Everything here is in US dollars. A sterling investor's world equity drawdowns and sleeve returns would both differ by whatever the pound did in the same window. The equity windows are also defined on equity. A crisis that arrives first in bonds, in credit or in cash is a different test, and this sample was not built to run it.

Finally, the monthly rounding. Each window includes a little of the run-up and a little of the recovery. On a daily series the equity falls are deeper than the monthly figures shown beside the sleeve results, and the trend returns would differ too.

What would change the conclusion

Three things, and none of them is the next crisis by itself.

The first is speed. The rule needs a move that has already been running long enough to be detected. The 2020 window ran from 12 February to 23 March, the shortest in the sample, and it produced both the worst result for the equity leg, -46.22%, and one of the two smallest gains overall at 10.50%. AQR's own simulation missed the 1987 crash for the same reason. If equity falls keep compressing into weeks rather than quarters, the record above overstates what the strategy has left to give.

The second is the equity leg. It's the leg the strategy is sold on and the leg that lost money in half the sample, including 46.22% in 2020. If it keeps failing in crises while fixed income carries the load, then what you're actually holding is a bond and commodity trend position wearing an equity hedge label, and it should be sized as one.

The third is the fee gap. A strategy that clears cash by 1.40 points a year after costs, as the live fund did over the ten years to 2025, is a different proposition from one that clears it by the 11.12% the gross factor produced over 431 months. Watch that spread rather than the backtest. If net returns to real funds settle near the 85 basis points the 2008 NBER paper found for 1994 to 2007, the crisis protection is still there and you're paying the whole of it away for the rest of the cycle.