Time-Series Momentum

Time-series momentum (often called absolute momentum) is the simplest trend rule there is: look at how an asset has done over the past several months, and if it's up, own it; if it's down, short it. There is no comparison to other assets and no view on whether the price is "fair." Each market is judged only against its own recent history. This is the quantitative heart of Trend Following, and it is momentum's close cousin, the difference being what you compare each asset to.

The other flavour, Cross-Sectional Momentum, ranks a whole universe against each other and buys the leaders. Time-series momentum skips the ranking entirely. A market can be a "winner" against itself (up over the year) even while it is the worst performer in its group — and time-series momentum will still go long. That independence is why the same rule works across stocks, bonds, currencies, and commodities all at once, and why a diversified trend book holds dozens of unrelated markets.

The rule

For each asset you compute a trailing return — a common choice is the past 12 months — and take its sign. Positive means long, negative means short. Then you scale the position so every market contributes the same amount of risk:

$$w_{i,t} = \text{sign}\big(r_{i,\,t-12 \to t}\big)\times \frac{\sigma_{\text{target}}}{\sigma_{i,t}} .$$

Here $w_{i,t}$ is the position weight in asset $i$, $r_{i,\,t-12\to t}$ is that asset's return over the last twelve months, $\sigma_{i,t}$ is its recent volatility (how much it typically wobbles), and $\sigma_{\text{target}}$ is the risk budget you want each position to carry. The sign decides direction; the ratio decides size. A calm market gets a bigger position than a jumpy one so that both add equal risk — this is Vol Targeting baked directly into the signal.

Worked example: sizing two trades

Suppose gold returned $+18\%$ over the trailing year and crude oil returned $-22\%$. The signs give the directions immediately: long gold, short crude. Now size them to a $10\%$ risk budget each.

• Gold has a recent volatility of $15\%$. Its weight is $\sigma_{\text{target}}/\sigma = 10\%/15\% = 0.67$ — a long position of about two-thirds of your unit notional.
• Crude is far jumpier at $40\%$ volatility. Its weight is $10\%/40\% = 0.25$ — a short of only a quarter unit.

Both positions carry the same $10\%$ risk even though crude gets a much smaller notional. That is the whole point: the sign chose the side, and the volatility scaling made the calm market the bigger bet so neither one dominates the book's swings.

Why the "smile" matters

Because trend rules cut losers and ride winners, they tend to be already positioned the right way when a slow crisis unfolds — long bonds and short equities as a sell-off grinds on, for example. That gives time-series momentum a convex, straddle-like payoff: strong in sustained moves up or down, a mild drag when markets go nowhere. It is one of the few systematic strategies that has historically done well in prolonged equity bear markets, which is why it is prized as a diversifier.

Where it stumbles

• Whipsaws. In choppy, directionless markets the sign flips back and forth, so you buy high and sell low repeatedly — the small central "loss" in the smile. This is why trend bleeds in calm range-bound years.
• Turnover and costs. Flipping positions and constantly re-sizing to volatility racks up trading, so the gross edge can shrink badly after costs.
• Look-back tuning. A 12-month window is a convention, not a law. Tuning it (say to 9m or 7m) to fit a backtest is textbook overfitting; robust books blend several look-backs.
• Sudden reversals. The convex payoff has a mirror image: sharp V-shaped rebounds — the market crashing then snapping straight back — are the worst environment, because trend gets caught short right as it turns.