Momentum
Yes — more than almost anything else in the factor zoo. Buying the past year’s winners (skipping the most recent month) and shorting the losers, the effect Jegadeesh & Titman published in 1993, has held up out-of-sample and across countries and asset classes ever since. It’s one of a small handful of anomalies on this site that keeps clearing an honest significance bar. Its real weakness isn’t decay — it’s crash risk: sudden, violent reversals when beaten-down stocks snap back all at once.
Jegadeesh & Titman (1993, Journal of Finance) measured a 12–2 momentum strategy over 1965–1989 at roughly 12%/year with a t-statistic of 4.3. This site rebuilds it value-weighted from the Kenneth R. French Data Library’s momentum factor and checks the same three windows it checks for every anomaly: the paper’s own sample, everything after publication, and the full history since — with Newey–West t-statistics to account for the autocorrelation in monthly returns.
Momentum comes out as one of the site’s clearest survivors, alongside industry momentum, betting-against-beta and quality-minus-junk, while most value definitions, accruals and the small-cap size effect decayed sharply or never cleared the higher bar that the sheer number of tested factors demands.
Momentum’s weak point isn’t fading — it’s occasional violent reversals. In 1932, 2009 and 2020, the stocks momentum had been shorting (recent losers) rallied hard all at once coming out of a market bottom, wiping out years of steady gains in a matter of months. That tail risk is the honest price of an otherwise durable signal.
At the textbook 1-in-20 threshold (t ≈ 2), easily. But that threshold assumes one test, and finance has run far more than one test. Harvey, Liu & Zhu (2016) argue the honest bar, once you account for how many predictors have actually been tried, is closer to a t-stat of 3. Momentum’s original 4.3 clears that stricter bar comfortably — unlike most of its zoo neighbors.
Check the t-stat yourself
The multiple-testing explorer lets you set your own assumption for how many signals have been tested, watch the significance threshold move, and see which of 18 replicated factors — including momentum — stay above it.
Open the explorer →Less likely than most of the factor zoo. Its original t-statistic (4.3) clears even the stricter ~3.0 threshold that accounts for how many other predictors have been tested, and the effect has held up out-of-sample and internationally for three decades — the profile of a real effect, not a fluke.
A sharp, fast reversal where the stocks momentum is shorting — recent losers — rally hard all at once, usually coming out of a market bottom. It happened in 1932, 2009 and 2020, each time erasing a big chunk of the strategy's cumulative gains in a few months.
Not lately. Built the same way from the 30 French industry portfolios (rank on trailing 12-month return, skip the most recent month, go long/short), industry momentum has been much weaker as a standalone signal since the late 1990s, even though Moskowitz & Grinblatt originally argued it explained most of individual-stock momentum.
Educational replication, not advice. Figures reflect this site’s value-weighted momentum construction and move as the data updates.