Two stocks rallied 30% last quarter. One sits in the top decile of its sector by relative strength. The other delivered the same return while half its sector did better. A trader treating both as “momentum” will hold them the same way and be wrong about one of them. The gap between them has a name in the literature: time-series vs cross-sectional momentum. It changes how you screen, how you size, and how long you hold.
How time-series momentum is measured
Time-series momentum asks one question of one asset: did this asset’s own past return predict its own future return? The textbook construction is a trailing 12-month total return with the most recent month skipped, often written as t-12 to t-1. If that return is positive, the asset stays in the long sleeve. If it flips negative, the position comes off. There is no peer comparison anywhere in the calculation. Every name in the universe gets evaluated against itself.
I size my time-series sleeve on exactly that 12-1 window because the short-term reversal effect inside the most recent month muddies the signal. Stocks that ran hardest in the last 20 sessions tend to mean-revert before they trend again. Skipping the t-0 month captures persistence without inheriting the reversal. The output is binary at the position level (in or out) and continuous at the portfolio level (total long exposure across all names with positive 12-1 returns).
A common operational mistake is to read time-series momentum as a single number. It is not. You compute it independently for every symbol. The S&P 500 having a positive 12-month return tells you nothing about whether AAPL has one. The signal is per-name, always.
How cross-sectional momentum is measured
Cross-sectional momentum asks the opposite question. Of all eligible assets, which ones outperformed their peers over the lookback window? The construction is a rank. You compute trailing returns for every name in your universe, sort them, and isolate the top decile (or top quintile, or top 30, depending on the sleeve size). Long the winners, sometimes short the losers, rebalance monthly.
What gets passed in is a return distribution. What comes out is a ranking. The absolute size of the return is irrelevant. A stock up 8% in a quarter when the median peer is up 2% ranks higher than a stock up 15% when the median peer is up 20%. The first is a leader. The second is a laggard wearing a green tag.
I sort my screener output by trailing 6-month return rank and drop everything below the 70th percentile of the eligible universe before any other filter runs. The numerical value of the 6-month return never enters the sizing rule. Only the rank does. That is the operational signature of cross-sectional thinking: ordinal data drives the decision, not cardinal.
Why RSI is not relative strength
This is where the conflation does real damage. Most traders meet the word “momentum” through RSI, and most quietly assume RSI is a relative-strength tool. It is not. RSI measures the magnitude of an asset’s own recent gains divided by the magnitude of its own recent losses, normalized to a 0-100 scale. There is no peer in the formula. Nothing in RSI compares AAPL to MSFT or to the index. It is a time-series oscillator with a misleading name. The “relative” in Relative Strength Index is the ratio of up-moves to down-moves within the same series, not the relation of one asset to another.
Genuine relative strength, the cross-sectional kind, is a ratio chart or a rank: stock divided by index, or stock rank within sector. When traders run an RSI screen and call it a momentum screen, they are filtering on time-series persistence and labelling the result “the strongest names”. Often the names that pop out are the most overbought, not the strongest relative to peers. The two screens overlap by accident at best. A trader using this conflation might exit a true relative-strength leader because its RSI reads above 80, while holding a peer-lagging name because its RSI reads in a comfortable middle band.
Different persistence properties
The two signals decay differently, which matters more than the labels do. Time-series momentum tends to persist over 3-12 month windows and reverses inside windows shorter than a month. Cross-sectional momentum persists too, but the rank is unstable: a stock can hold the top decile for six months without ever moving more than a percentage point relative to the rank below it. Small changes in absolute return swap the order constantly, even when the underlying trend has not changed.
This is why time-series sleeves rebalance less often than cross-sectional ones. If the 12-1 return is positive, that signal does not flip casually. A cross-sectional rank can flip on a single 5% move in a single peer. Treating the two as the same signal leads to either over-trading the time-series book or under-trading the cross-sectional one. The rebalance cadence has to follow the half-life of the signal, not the calendar.
Time-series vs cross-sectional momentum in practice
A practical use of both is to layer them. Cross-sectional rank as the universe filter. Time-series sign as the position-on rule. Step one, rank every eligible name on trailing 6-month return and keep the top quintile. Step two, inside that subset, require a positive 12-1 return. The first step says “this name is a leader against its peers”. The second says “this name’s own trend is still alive”. Names that fail step one are not in the conversation. Names that pass step one but fail step two are leaders coming off their highs, and they sit in cash.
This is the approach a trader using a sector rotation framework might extend down to the single-name level: first find the leading sector, then find the leading stocks inside it, then require time-series confirmation on each one before deploying capital. The same logic underwrites the network-momentum lead-lag research line, which formalizes the idea that the cross-sectional structure and the time-series signal carry distinct information that compounds when combined.
When time-series momentum fails
Time-series momentum is not a trend filter and not a regime filter. It will hold a name that has rolled over slowly across 11 months because the 12-1 return is still nominally positive. It will exit a name in a sharp single-month correction even when the larger trend is intact, because the calculation crossed zero. Treating a positive 12-1 return as “the trend is up” is the textbook misread. The signal is a binary on a backward-looking sum. It tells you nothing about volatility regime, nothing about the slope of the recent advance, and nothing about whether the name is leading its peers.
The second misread is treating a monthly rebalance as fast enough. Inside a high-volatility regime, a 12-1 return can flip sign and flip back inside a single rebalance window. The practitioners I respect in the Paul Tudor Jones lineage pair the time-series signal with a volatility-scaled position size precisely because the on/off binary leaves all of the path-dependency on the table. The signal tells you whether to be involved. It does not tell you how big.
When cross-sectional momentum fails
The cross-sectional rank fails in two specific situations, and both are non-obvious. The first is a flat distribution. When the spread between the top decile and the bottom decile collapses, the rank is statistically meaningless. You are still holding the “winners” but the difference between winner and loser is noise. The second is a falling-tide market. Cross-sectional rank says “rank stocks against each other”. It does not say “rank stocks against cash”. In a -25% market, the top-decile name is still in the top decile when it is down 18%. The signal works, the trade does not.
The gap closes when a rate-of-change time-series overlay sits on top of the cross-sectional rank. The cross-sectional book identifies the strongest names. The time-series filter answers the separate question of whether any of them have a positive return at all. A trader running cross-sectional alone in 2008 was long the strongest decliners and learned the lesson the expensive way.
Two signals, two sleeves, one strategy
The cleanest mental model is that time-series momentum tells you whether to be in a name and cross-sectional momentum tells you which name to be in. The first is a participation rule. The second is a selection rule. They were never the same thing, even when traders call them both “momentum” and run them through the same RSI window. The names that survive both signals at the same time are the ones worth sizing up. The names that fail one but pass the other are the source of most of the noise in a “momentum strategy” that does not separate the two.
Learn the pattern. Ride the trend. Keep the gains.
Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.