Cross-Sectional Momentum in Equities — the strongest documented anomaly, and how much of it survives costs

Buying past winners and shorting past losers over 3-12 month horizons produced the strongest and most-replicated anomaly in the original Jegadeesh-Titman (1993) study. The effect is real in the original sample and holds up internationally, but it decayed roughly 5x post-1990s, crashes violently in market-rebound states, and after realistic trading costs a meaningful slice of its edge — especially in small caps — disappears.

Jegadeesh-Titman 1993 original numbers

- Sample: NYSE and AMEX stocks, CRSP daily returns file, January 1965 to December 1989 (25 years). - 16 formation/holding combinations tested (3, 6, 9, 12 months each, crossed) — all 16 produced positive average returns, ranging 0.9% to 1.3% per month across combinations. - Best combination: 12-month formation / 3-month holding → 1.49% monthly for the winner-minus-loser portfolio (≈17.88% annualized, arithmetic). - 6-month formation / 6-month holding: 12.01% annualized excess return. - 6-month momentum, full 1965-1989 sample: ~1% monthly, t-statistic 3.07 — the number this wiki treats as the headline significance figure, since it clears the t>3 bar discussed in Multiple Testing: Why t>1.96 Is Not Enough — the bar this wiki uses to grade a factor's significance.

International/long-horizon check

momentum has been tested across roughly 40 countries with up to 150 years of historical data, and an independent backtest going back to 1927-2013 shows 8.3% indicative annualized performance — i.e. the original 1993 US result is not a single-sample artifact, though the exact effect size varies a lot by period (see below).

What does NOT hold up

the ~10-12% annualized figure from the original sample. Momentum returned roughly 10% annually in the 1990s, declining to closer to 2% in recent periods. Momentum profits stayed positive through the 1990s right after publication (evidence against pure data-snooping in the original study), but the literature documents:

- ~50% of the momentum alpha disappearing post-publication, consistent with the general post-publication decay pattern covered in Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real. - Momentum profits turning insignificant in the late 1990s in certain sub-periods, with alpha declining sharply again in 2007-2015. - A separate reversal pattern: momentum returns show significant reversals 4-5 years after formation — the winner-loser spread that built up over the formation/holding window partly unwinds on a multi-year horizon.

Momentum crashes

Momentum is not a smooth premium — it has a small number of catastrophic months concentrated in specific market states.

- Full-sample maximum drawdown: -87.41%. - March 2009 = 7th worst momentum month; April 2009 = 4th worst in the sample; 3 of the 10 worst momentum months ever occurred in 2009, a three-month window when the market rallied sharply and volatility fell. - Mechanism: momentum crashes occur in "panic" states — following market declines, when volatility is high — and hit contemporaneously with market rebounds. Crash losses are mostly attributable to the short side: past losers (the stocks momentum shorts) rally sharply during the rebound. - Because the trigger condition (prior decline + high vol) is observable, the 2009 crash is described as partly forecastable — this is the empirical hook for any risk-managed/vol-timed variant of the strategy.

What does NOT work as stated

naive constant-exposure long-short momentum held through a volatility spike after a crash — that is exactly the setup that produced the -87% drawdown and the worst months on record.

Transaction cost survival threshold

This is the number that decides whether the strategy is usable, not just documented — see Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one and Computing a Strategy's Transaction-Cost Threshold — the four-step check that decides whether a documented edge is tradable for the general method; here are the momentum-specific figures.

- Turnover: momentum runs 80-90% average one-sided annual turnover — high relative to value/quality strategies, which is why costs bite harder here. - AQR live-fund data, 2009-2016: average trading cost 23 bps, broken down by segment: US large-cap 12 bps, US small-cap 32 bps, international 25 bps. - Net of costs: gross momentum returns of 1-1.5%/year annualized minus 23 bps of costs leaves ≈0.77-1.27% net annualized — positive but far below the 1990s headline figure, and this is *after* the post-1990s decay already described above, not instead of it. - Small-cap momentum is often unprofitable after costs: one small-cap fund underperformed its benchmark by 1.2% despite positive gross returns, consistent with the 32 bps small-cap cost figure eating the whole edge. - Capacity limit: as much as $5 billion of flow into a given momentum strategy is enough to make the apparent profit opportunity vanish — a crowding constraint independent of the per-trade cost number. - Robustness check: AQR's research finds costs would need to increase more than fivefold from 2009-2016 levels to eliminate momentum returns entirely — i.e. the surviving net premium is not on a knife's edge at current cost levels, even though small-cap implementations already are. - Long-only implementations are more resilient than long-short after costs (no borrow cost, lower turnover on the short leg is simply absent). - A tax-aware implementation (launched 2012) cut the *effective* tax rate from 11.4% to 5.3% by holding winners longer and selling losers faster — a portfolio-construction lever separate from the trading-cost number, relevant to anyone trying to push net returns back toward the gross figure. Most of that 6.1-point drop is not federal tax: the federal component only improved from 1.2% to 0.5% (0.7 points), so the bulk of the gain comes from non-federal parts of the effective rate — don't read this as "federal tax drag fell 6 points."

Related

- Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real — momentum is the reference case for post-publication decay itself: the ~50% alpha-disappearance figure here is one of the concrete data points that general page draws on. - Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one and Computing a Strategy's Transaction-Cost Threshold — the four-step check that decides whether a documented edge is tradable — the 23/12/32 bps figures above are a worked instance of the cost-threshold method those pages describe generically. - Time-Series Momentum in Futures — a backtest with Sharpe near 1.0 that live CTAs never matched — the sister momentum effect in futures; useful contrast for whether momentum survives costs better outside single-name equities. - Multiple Testing: Why t>1.96 Is Not Enough — the bar this wiki uses to grade a factor's significance — the 6-month t-stat of 3.07 reported here is the specific number to check against that page's t>3 bar when judging whether cross-sectional momentum clears today's stricter significance standard.