Trend Following, Finally Explained: Time Series Momentum

Managed futures funds, the "trend followers," have existed for decades. Their pitch is almost insultingly simple: buy things that have been going up, sell things that have been going down, across every market they can trade. Academics mostly regarded this as astrology with a Bloomberg terminal.

In 2012, Tobias Moskowitz, Yao Hua Ooi and Lasse Heje Pedersen ran the trend follower's rule across 58 liquid futures markets and found that it worked. Not in a few markets. In every single one of them. The paper legitimised an entire industry and gave the effect its name: time series momentum.

The problem: the momentum we knew about was the wrong kind

The famous momentum effect, from Jegadeesh and Titman, is cross-sectional. It compares stocks to each other: buy the ones that outperformed their peers, short the ones that lagged. It is fundamentally a relative bet, and it is market-neutral by construction. If every stock rises 20%, cross-sectional momentum has nothing to say.

But that is not what a trend follower does. A trend follower does not care how oil is doing relative to copper. It asks a much simpler question about each market separately: has this thing been going up? If yes, be long. If no, be short. It is an absolute judgment, one market at a time, and it will happily be long every single market at once if everything is rising.

Nobody had rigorously tested that version. Cross-sectional momentum had a large literature. The thing practitioners actually did had none.

The key idea via analogy: does the past predict the future for each market on its own?

Their test is beautifully clean. For each of 58 futures contracts (stock index futures, currencies, commodities, bond futures), do the following:

1. Look at how the market performed over the past 12 months.
2. If the return was positive, go long. If negative, go short.
3. Size the position so that each market contributes roughly the same amount of risk, scaling down the wild ones and up the calm ones. This volatility scaling matters enormously in practice: without it, a couple of volatile commodities would dominate the whole portfolio.
4. Hold for a month, then repeat.

The results:

• The signal was positive in all 58 markets. Every single one showed a tendency for the past 12 months to predict the next month in the same direction. Fifty-eight independent markets all pointing the same way is not a fluke.
• The persistence lasts roughly 1 to 12 months, then partially reverses over longer horizons. Trends run, and then they overshoot and give some back. That specific shape, under-reaction followed by delayed over-reaction, is exactly what behavioural theories of sentiment predict.
• A diversified portfolio of these strategies across all asset classes delivered substantial abnormal returns with little exposure to the standard asset pricing factors. In other words, it is not disguised market beta or a hidden value tilt.
• It performed best in extreme markets, at both ends. When markets go badly wrong, trends tend to be long and persistent, and a trend follower is already positioned short. This is the property that makes the strategy so attractive to institutional investors: it has historically been a friend when everything else is a problem.

Why it mattered

• It gave managed futures an academic foundation. An industry running many billions of dollars had, until this paper, no serious peer-reviewed evidence base. Afterwards it had one, and allocations followed.
• It reframed trend following as a risk premium, not a trick. The paper connects the effect to the behaviour of the actual participants in futures markets. Speculators (trend followers) tend to be positioned with the trend, and hedgers (commercial producers and consumers) against it. That is consistent with speculators being paid to take risk off hedgers' hands, a story that goes back to Keynes.
• It explained the crisis performance. Trend followers had a famously good 2008, and this paper explains why that was structural rather than lucky: the "long crisis" profile falls out of the strategy's construction.
• It unified the field. Cross-sectional and time-series momentum are related but distinct. Showing both exist, and that the time-series version works across every asset class, made momentum look less like an equity oddity and more like a fundamental feature of how prices adjust.

The honest limitations

• The good years were largely before the paper. Trend following has had a noticeably harder time in much of the period since publication, particularly the long, choppy, central-bank-driven markets of the 2010s. Whether that is crowding, regime change or bad luck is genuinely contested.
• It relies on getting the risk scaling right. The volatility targeting in step 3 is not a detail, it is load-bearing. The strategy's attractive properties depend heavily on how you size positions, and that gives a lot of room for a backtest to be flattered.
• Whipsaws hurt. In range-bound markets, a trend follower buys every false breakout and sells every false breakdown, bleeding steadily. The strategy has long, demoralising flat periods, which is precisely when investors abandon it.
• The explanation is still debated. The paper leans toward a behavioural under-reaction story supported by a hedging-pressure argument. Both are plausible. Neither is settled, and a strategy without a firmly understood cause is a strategy you cannot be sure will keep working.

The one-line takeaway

Moskowitz, Ooi and Pedersen tested the trend follower's oldest rule, buy what has gone up over the past year and sell what has gone down, across 58 futures markets, found it worked in all of them, and in doing so turned an industry's folk wisdom into a documented cross-asset risk premium.