Cross-sectional momentum is not a chart signal. It is a ranking rule: compare a defined set of instruments by a pre-set past-return measure, then turn those ranks into a portfolio rule.
That makes it more testable than “this market looks strong.” It also makes the hidden choices impossible to ignore. The universe, membership history, ranking date, rebalance timing, costs and constraints can each change the result.
The relative-strength idea
Cross-sectional momentum asks which members of a group did better or worse than the other members over a stated past interval. The original Jegadeesh and Titman study examined relative-strength portfolios; the Kenneth R. French Data Library makes the same basic idea inspectable through portfolios formed on prior returns (Jegadeesh and Titman, 1993, French Data Library: momentum portfolios).
The important word is relative. A market can have fallen and still rank near the top of its chosen group. Another can have risen and still rank near the bottom. The rule compares peers; it does not declare that an instrument is attractive in isolation.
defined universe -> calculate the same past-return measure for every member
-> rank the members on one fixed date
-> map ranks to a pre-defined portfolio rule
-> hold / rebalance under stated execution assumptionsIt is not the same as trend following
Trend following, or time-series momentum, asks whether one instrument's own return is positive or negative over a chosen lookback. Cross-sectional momentum instead asks where that instrument sits in a ranking against peers. The two can agree, but they answer different questions; the French library separately documents portfolios built from prior-return ranks and portfolios split by size and momentum (French Data Library: prior-return deciles, French Data Library: size and momentum).
| Question | Time-series trend following | Cross-sectional momentum |
|---|---|---|
| Comparison | An instrument against its own past | An instrument against its peers |
| Output | Often a directional state | A relative rank or rank bucket |
| Essential input | One price series and its rule | A universe, comparable histories and a ranking rule |
| Common hidden choice | Signal and exit timing | Universe membership and rebalancing |
For a trader looking at a single FX pair, there is no cross-section until a group is defined. That might be a stated set of currency pairs, indices, futures or equities. Mixing unlike instruments without first defining the comparison basis is not a neutral shortcut; it changes the hypothesis.
The research is a starting point, not a trading instruction
The academic record is why momentum is worth studying, not a licence to copy a chart label. Jegadeesh and Titman's paper is evidence from a particular research design, while the French library's published construction shows that the result depends on operational decisions such as eligible securities, prior-return data and the dates on which portfolios are formed (Jegadeesh and Titman, 1993, French Data Library: construction notes).
That caveat matters. A historical factor series is not your brokerage account. A live implementation has its own universe, spreads, commissions, slippage, borrow availability where short positions are involved, trading hours and position limits. It may also face the ugly periods that a smooth rank chart does not prepare you for.
The criticism is not that ranking is subjective. It can be written precisely. The criticism is that a precise rank can still rest on convenient data selection or a selection of lookbacks, asset groups and rebalance rules made after inspecting results. Why survivorship bias flatters a backtest is relevant here: a universe is historical data, not just a list typed into a platform.
How you'd actually test it
Do not test “whether momentum works.” Test one frozen portfolio construction. The French examples are useful precisely because their construction is written down; mirror that discipline in your own test (French Data Library: momentum construction, French Data Library: size-and-momentum construction).
| Part of the rule | What must be fixed before results |
|---|---|
| Universe | Eligible instruments, inclusion history, currency treatment and delistings or symbol changes |
| Ranking input | Return definition, observation window, price field and any skip period |
| Portfolio map | Which rank buckets are held, long-only or long-short treatment, weights and concentration limits |
| Clock | Ranking timestamp, rebalance schedule, order timing and timezone |
| Execution | Bid-ask spread, commission, slippage, financing, liquidity assumptions and short constraints |
| Evaluation | A baseline, untouched out-of-sample data and the whole distribution of returns and drawdowns |
Run the test with point-in-time membership. Keep the instruments that later disappeared and record substitutions instead of quietly replacing them. Then compare nearby, sensible specifications without crowning the best historical cell. Out-of-sample testing and trade frequency in a prop backtest cover the two checks that a ranking model often evades: whether it survives unseen data and whether turnover leaves anything after friction.
realbacktesting is a trading-software studio for cTrader built around verifiable work: its published cBot figures are cTrader-native backtests that readers can reproduce in their own cTrader. The standard here is the same. Make the rank rule and the historical universe reproducible before giving the result any authority.
Frequently asked
Is cross-sectional momentum the same as a momentum indicator?
No. A momentum indicator normally transforms one instrument's price history. Cross-sectional momentum compares the same pre-defined calculation across several instruments and uses their relative ordering.
Can a cross-sectional momentum rule be tested on one instrument?
Not as a cross-sectional rule. One instrument can support a time-series rule, but relative ranking requires a defined peer group at each ranking date.
Does published momentum research guarantee a live result?
No. A published result belongs to its data, construction and assumptions. A live or platform-specific implementation needs its own test with its own costs, constraints and out-of-sample evidence.
The stubborn takeaway
Cross-sectional momentum earns its name only when the peer group is as explicit as the rank. If the universe changes whenever the result looks inconvenient, the model is not systematic; it is hindsight with a spreadsheet.
How it is used
<the mechanics: structure, rules, a worked example. Use a table or a ```text diagram where it clarifies.>
What the evidence and the critics say
<honest and balanced: subjectivity, hindsight bias, weak or mixed empirical evidence where that is true. Cite real sources.>
How you'd actually test it
<the objective rules you'd need to backtest it — tie to realbacktesting's "test it, don't trust it" ethos. Link a relevant blog explainer.>
Takeaway
<a stubborn, memorable line that ties back to testability. No advice, no product pitch.>