Momentum

Definition

Momentum strategies buy recent winners and avoid or short recent losers, on the premise — among the most-replicated in finance — that intermediate-horizon relative performance persists. Two distinct forms: cross-sectional (rank instruments against each other; Jegadeesh-Titman) and time-series (each instrument against its own past; Moskowitz-Ooi-Pedersen — the engine of strategy-futures-trend-following).

How it works / structure

• Cross-sectional recipe: rank universe by trailing 12-month return excluding the last month (the 12-1 convention — the skipped month avoids short-term reversal contamination, strategy-mean-reversion); hold the top decile/quintile; rebalance monthly.
• Time-series recipe: long where trailing N-month excess return is positive, short/flat where negative; volatility-scale positions (risk-volatility-targeting).
• Parameters (engine-executable): lookback window (3-12m), skip window (0-1m), holding/rebalance period, universe, long-only vs long-short, volatility scaling on/off, crash filter (see below).
• Platform bindings: trend_state, range_52w_pos_pct (proximity to 52-week highs is a related documented signal).

When it applies

Diversified universes (single-name momentum is factor exposure plus idiosyncratic noise), intermediate horizons (weeks to months — momentum inverts at both shorter and longer horizons), and regimes without sharp reversals. Long-short versions need borrow capacity (ms-short-locate-borrow, strategy-short-selling).

Risk profile & failure modes

• Momentum crashes: Daniel-Moskowitz (2016) documented rare, violent losses concentrated in market rebounds after panics — the short-losers side rallies hardest; 1932 and 2009 are the canonical episodes. Crash filters (volatility-conditioned scaling) mitigate at cost.
• Turnover and costs: monthly-rebalanced momentum is high-turnover; realistic friction (ms-slippage-friction) consumes a large share of the paper premium.
• Post-publication decay: McLean-Pontiff (2016) measured meaningful attenuation; current-period edge is an open question, never an assumption.
• Crowding: momentum unwinds are correlated across managers — drawdowns cluster with peers (port-correlation-budgets).

Evidence & limits

Jegadeesh-Titman (1993) found 12-1 winners outperformed losers by roughly 1% monthly in their 1965-1989 US sample; the effect has been replicated across countries, asset classes, and centuries of data (Moskowitz-Ooi-Pedersen 2012 for time-series across 58 futures markets). It remains debated as risk premium vs behavioral underreaction, decays post-publication, and carries the documented crash profile. Momentum is the best-evidenced strategy family in this KB — and still not a guarantee.

Falsifiable-thesis examples

Illustrations only, not signals:

• “The top-decile 12-1 momentum portfolio in universe U will beat the universe average over the next 6 months” — falsified by the realized relative return.
• “X, in the top 5% of its 52-week range, will outperform its sector ETF over the next quarter” — falsified by the pair’s returns.

Cross-references

• Time-series twin: strategy-futures-trend-following
• Opposite horizon effects: strategy-mean-reversion
• Risk shaping: risk-volatility-targeting, regime-volatility, risk-max-drawdown-budget
• Method discipline: lens-quantitative (multiple-testing, decay)