TL;DR
Over horizons of roughly 3–12 months, stocks that recently outperformed keep outperforming and recent losers keep lagging. A long-winners / short-losers portfolio has historically earned on the order of 1% per month, and the pattern appears in nearly every market and asset class — one of the most robust anomalies in finance. The catch: momentum is crash-prone, high-turnover, and concentrated in the hardest-to-arbitrage corners of the market.
A 30-year arc
Jegadeesh & Titman (1993) — the founding result: buying past 3–12-month winners and shorting losers earns ~1%/month over the next 3–12 months, unexplained by market risk.
Chan, Jegadeesh & Lakonishok (1996) — price momentum and earnings-surprise drift (PEAD) each predict returns, additively (the strongest sort ~1.74%/mo). Frames momentum as underreaction to information.
Rouwenhorst (1998); Asness, Moskowitz & Pedersen (2013) — momentum is pervasive: across international equities and, in Value and Momentum Everywhere, across asset classes (with value as its negative correlate).
Moskowitz & Grinblatt (1999) — much of individual momentum is really industry momentum.
George & Hwang (2004) — nearness to the 52-week high forecasts returns better than past returns (~1.06%/mo vs JT 0.38% / MG 0.25%) and does not reverse.
Jegadeesh & Titman (2001) — profits persist out of sample; only a partial reversal appears, at years 4–5.
Novy-Marx (2012) — momentum is "echo": the intermediate horizon (months t-12 to t-7) drives it, not the most recent months.
Blitz, Huij & Martens (2011) — residual (factor-neutral) momentum is smoother and less crash-exposed.
Moskowitz, Ooi & Pedersen (2012) — time-series (trend) momentum: an asset's own past return predicts its future, across futures markets.
Daniel & Moskowitz (2016) — momentum crashes: rare, severe drawdowns in panic-rebound markets.
Avramov, Chordia, Jostova & Philipov (2007) — momentum lives almost entirely in low-credit-quality firms (under 4% of rated market cap).
Sub-threads (the momentum family)
Cross-sectional price momentum (Jegadeesh-Titman) · earnings momentum / PEAD · 52-week-high · intermediate / echo · residual · industry · time-series / trend · cross-asset. They overlap heavily but are not identical — a momentum-aware book treats them as one correlated complex.
Why it works
The consensus channel is behavioral underreaction: investors and analysts absorb news gradually. Hong & Stein model gradual information diffusion; Daniel, Hirshleifer & Subrahmanyam attribute initial underreaction (then delayed overreaction) to overconfidence and self-attribution; Barberis, Shleifer & Vishny invoke conservatism and representativeness. Risk-based explanations exist but struggle to span both momentum's magnitude and its crash profile.
The dark side
Crashes — momentum suffers rare, severe losses when beaten-down stocks rebound (1932, 2009); the unconditional Sharpe hides a fat left tail.
Turnover & costs — monthly rebalancing makes momentum one of the most cost-sensitive factors (see the cost-of-trading mission).
Concentration & capacity — profits cluster in small, illiquid, low-rated names that are hard to short, limiting capacity.
Does it survive out of sample?
McLean & Pontiff (2016) find published anomalies decay ~30–58% post-publication; momentum is among the more durable and remains positive out of sample. Our replications recompute the canon from building blocks and score it on the holdout — see Jegadeesh-Titman cross-sectional momentum, industry momentum, time-series momentum, and asset-class momentum.
Run it yourself
Curriculum — Mission 1 uses a planted momentum signal to teach the overfitting trap; the Lab's synthetic market ships mom_1m / mom_3m / mom_12m factors.
Playground — prototype a momentum signal in the browser.
Competitions — submit a momentum strategy and see it scored out of sample.
