Time-Series Momentum (Futures)

Thesis (edge)

Time-series momentum is the simplest trend rule that exists. You look at one market, ask whether it went up or down over the past year, and you take a position in that same direction for the next month. There is no comparison against other markets and no ranking. Each market votes only on itself.

The behavioural story is that information gets into prices slowly. When something changes for oil or for the bond market, some investors react at once, some react after they see the first move, and some react only once the story is on the front page. That staggered reaction stretches a piece of news into a drift that lasts months rather than a single clean jump.

There is a second story that has nothing to do with mistakes. Futures markets exist so that producers and hedgers can pass risk to someone else. Speculators who take the other side get paid for holding that risk, and the direction they need to lean often lines up with the recent trend. Under this reading, trend profits are partly a fee for providing liquidity to people who need to hedge, which is a reason the edge can persist even though everybody knows about it.

Where it works (regimes)

It works when moves are big and slow: a commodity supply shock that takes a year to resolve, a rate hiking cycle, a currency that grinds in one direction for eight months. 2008 and 2022 are the showcase years, because bonds, energy and currencies all moved a long way in one direction and a trend book was already positioned.

It struggles when markets are quiet and directionless. A range-bound year produces a string of small losses as positions flip from long to short and back. It also struggles at sharp reversals: because positions are scaled up when a trend has been running, the book tends to be at maximum size right when the trend breaks. March 2020 hurt many trend books for exactly this reason, before later paying them back.

Signals

• The core signal is the sign of the market's own return over the past 12 months. Positive means long, negative means short.
• Some versions blend several lookbacks, for example 3, 6 and 12 months, and average the resulting signals. This makes the position change more gradually and usually reduces turnover shocks.
• A common refinement is to use the sign only, not the size, of the past return. A market that rose 60 percent and a market that rose 4 percent both get a full-size long. This sounds crude but it stops one wild market from dominating the book.
• If you want fewer trades, add a deadband: stay flat when the trailing return is within a small band around zero.

Portfolio construction

Every market gets the same risk budget, not the same dollar amount. A bond future moves far less in percentage terms than natural gas, so it needs a much bigger notional position to contribute the same risk. Dividing by recent volatility does this job.

After sizing each market, scale the whole book so that expected portfolio volatility sits at your target, commonly around 10 percent annualised for a standalone programme. Diversification is doing a lot of the work here: 40 markets that trend independently give a far smoother line than 5 markets that all track the dollar.

Watch for hidden concentration. Several currency pairs against the dollar, plus gold, plus emerging market indices, can quietly become one big dollar bet. Grouping markets into sectors and capping sector risk is a cheap fix.

Risk model

The main risk is a correlated reversal, where the trend flips in many markets at the same time because one macro driver flipped. Vol targeting helps but does not save you, because vol targeting reacts to volatility rather than to direction.

Set a portfolio drawdown rule in advance, for example cut risk by a third after a 10 percent drawdown. Also watch realised correlation between sectors, since the effective number of independent bets can quietly fall from 20 to 4 in a crisis.

Costs & implementation

Turnover is modest at monthly rebalancing, which is the main attraction of this version over faster breakout systems. Costs are dominated by rolls rather than by signal-driven trades in the liquid markets.

The roll rule matters more than people expect. If you build the continuous series one way and trade another way, your backtest will show returns you cannot collect. Be explicit: roll on a fixed schedule, or roll on volume, but use the same rule in the simulation and in production.

Capacity is high in equity index, bond and major FX futures, and much lower in the agricultural and softs markets. If you plan to run size, test how the results change when you drop the 10 thinnest markets.

Failure modes

• Choppy markets producing repeated whipsaws and a slow bleed.
• Reversal risk at the top of a trend, when the position is largest.
• Fitting the lookback to history, then discovering that 12 months was lucky and 9 months was not.
• Backtesting on badly built continuous contracts, which can create signals that never existed.
• Assuming every listed future is tradeable at scale.

Our Notes & Suggestions

Do not treat the 12-month lookback as sacred. Run 3, 6, 9 and 12 months, and if only one of them works, be suspicious. A real edge should show up as a broad plateau across sensible parameters, not a single spike.

Blend speeds rather than picking a winner. An average of a fast and a slow signal typically loses a little in the best trending years and loses much less in the bad ones.

Finally, judge this strategy by what it adds to a portfolio, not by its own Sharpe ratio. A standalone Sharpe near 0.5 that happens to make money in equity bear markets can be more valuable than a Sharpe of 0.8 that dies in exactly the same months as your stocks.