The momentum anomaly is the empirical regularity that securities which have outperformed over an intermediate look-back window (roughly 3–12 months) tend to continue outperforming over the next several months, while past laggards continue to lag. It is one of the most robust and pervasive patterns documented in financial markets — it persists across decades, countries, and asset classes, and it survived the Fama-French three-factor model as a separate, additive factor. Its core tension: the effect is strong and durable in a diversified long/short portfolio, yet it is fundamentally a relative-performance/portfolio phenomenon, prone to rare but violent crashes, and it does not transfer cleanly to a concentrated, discretionary single-name swing trade.

Two distinct forms — keep them separate

• Cross-sectional momentum (the canonical "momentum factor"). Rank a universe of stocks by trailing return, go long the top winners and short the bottom losers. It is relative: a stock qualifies by beating its peers, even in a down market. This is the Jegadeesh & Titman (1993) result and the UMD/MOM factor.
• Time-series (absolute / "trend") momentum. Each instrument is judged against its own past return, not against peers: if an asset's own trailing 12-month return is positive, go long; if negative, go short. Documented by Moskowitz, Ooi & Pedersen (2012) across 58 futures contracts and four asset classes. This is the academic backbone of trend-following / CTA strategies.

The two are correlated but not identical — an asset can have positive absolute momentum yet rank in the bottom cross-sectional decile (and vice versa). A swing trader leaning on "relative strength" is invoking the cross-sectional form.

How it's measured / formed

The canonical Jegadeesh-Titman construction:

• Formation (look-back): trailing return over the past 3, 6, 9, or 12 months. The most-cited specification is "12-1" — the prior 12 months excluding the most recent month.
• Skip period: the most recent ~1 month (or 1 week in the original paper) is deliberately skipped to avoid contamination from the short-term reversal effect and bid-ask bounce (Jegadeesh & Titman 1993).
• Holding: typically 3–12 months, with periodic rebalancing.
• Portfolio: rank the universe, go long the top decile/quintile ("winners"), short the bottom ("losers") — the winners-minus-losers (WML) zero-cost spread.

J&T tested all 16 combinations of {3,6,9,12}-month formation × {3,6,9,12}-month holding; every combination produced positive average returns (Jegadeesh & Titman 1993). The 6-month/6-month version became the field standard in later research.

How it's used in practice

• Quant factor sleeves. Run as a diversified, periodically-rebalanced long/short (or long-only "tilt") with dozens-to-hundreds of names — the form in which the edge is actually documented.
• Relative-strength stock selection. Discretionary and CANSLIM-style traders buy names leading their peer group / the index (the retail expression of cross-sectional momentum).
• Trend-following (time-series). CTAs go long/short each market on the sign of its own trailing return, volatility-scaling positions.
• Combining with value. Because value and momentum are negatively correlated (≈ −0.5 to −0.6 across markets; Asness, Moskowitz & Pedersen 2013), a value+momentum blend diversifies strongly even though both legs are individually positive.

Standing & evidence

The momentum anomaly is among the best-evidenced in finance:

• Original result. Jegadeesh & Titman (1993) found WML profits of roughly ~1% per month over 1965–1989; the 12-month-formation / 3-month-holding spread averaged about 1.31% per month in their data.
• Survived factor models. Carhart (1997) added momentum (UMD, "Up-Minus-Down" — the prior-12-month winners-minus-losers spread) to the Fama-French three factors, creating the four-factor model; the factor was distinct from market, size, and value and improved explanatory power.
• Pervasive. Asness, Moskowitz & Pedersen (2013), "Value and Momentum Everywhere," documented positive, significant value and momentum premia across eight markets/asset classes with a common factor structure — strong evidence against pure data-mining.
• Absolute / time-series form. Moskowitz, Ooi & Pedersen (2012) found the past 12-month excess return positively predicts the next month's return for every one of 58 futures contracts examined (1985–2009), with a composite Sharpe near 1.28 vs ~0.38 for buy-and-hold.

The cause remains debated — behavioral under-reaction / delayed over-reaction (sentiment) vs. compensation for a real (time-varying) risk. The pattern itself is not seriously disputed; its explanation is.

Strengths & limitations

• Strength: durable, out-of-sample, cross-asset, and additive to value and the standard factors — a genuine, repeatedly-replicated edge at the diversified-portfolio level.
• Momentum crashes (the headline risk). Daniel & Moskowitz (2016), "Momentum Crashes," show the strategy suffers rare, severe drawdowns. Crashes cluster in "panic" states — after sustained market declines and amid high volatility — and hit precisely as the market rebounds (the beaten-down "loser" shorts spike). They report that 14 of the 15 worst momentum months occurred when the prior two-year market return was negative and the contemporaneous market return was positive. The loser leg behaves like a short option (negative skew). This is the single most important caveat.
• Portfolio ≠ single name (the #1 misuse). The documented edge is a diversified, market-neutral, dozens-of-names effect. A discretionary swing trader who buys one strong stock is exposed to that name's idiosyncratic risk and the crash dynamic, with none of the cross-sectional averaging that produces the academic premium. "Momentum works" does not license a concentrated bet.
• Skip the most recent month. Buying a stock that just spiked in the last week or two invokes the short-term (1-month) reversal, which runs opposite to intermediate momentum — recent sharp moves tend to mean-revert. The canonical signal excludes the most recent month for exactly this reason.
• Regime/turnover. High turnover (transaction costs erode the gross premium); the edge weakens or inverts in sharp reversals and is non-stationary.

Not the RSI indicator

"Momentum" here is the return-continuation factor above — a portfolio of past winners vs losers. It is a different thing from RSI (Wilder's Relative Strength Index), a bounded 0–100 oscillator typically used to flag short-term overbought/oversold mean-reversion. RSI is also distinct from relative strength (a stock's performance versus the index/peers), which is the discretionary cousin of cross-sectional momentum. Do not let the factor's strong academic standing be borrowed by the RSI oscillator, whose standalone edge is weak.

Sources

• Jegadeesh, N. & Titman, S. (1993), "Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency," Journal of Finance 48(1), 65–91 — bauer.uh.edu/rsusmel/phd/jegadeesh-titman93.pdf; onlinelibrary.wiley.com/doi/abs/10.1111/j.1540-6261.1993.tb04702.x
• Carhart, M. (1997), "On Persistence in Mutual Fund Performance," Journal of Finance — UMD/four-factor model (Wikipedia: Carhart_four-factor_model; cross-checked).
• Moskowitz, T., Ooi, Y.H. & Pedersen, L.H. (2012), "Time Series Momentum," Journal of Financial Economics — aqr.com/Insights/Research/Journal-Article/Time-Series-Momentum; w4.stern.nyu.edu/facdir/lpederse/papers/TimeSeriesMomentum.pdf
• Asness, C., Moskowitz, T. & Pedersen, L.H. (2013), "Value and Momentum Everywhere," Journal of Finance 68, 929–985 — aqr.com/Insights/Research/Journal-Article/Value-and-Momentum-Everywhere
• Daniel, K. & Moskowitz, T. (2016), "Momentum Crashes," Journal of Financial Economics 122, 221–247 — nber.org/papers/w20439; nber.org/system/files/working_papers/w20439/w20439.pdf
• Foxholm Financial summary of Jegadeesh & Titman (skip-week / reversal detail) — foxholm.com/q/research/jegadeesh-titman-momentum/