Momentum: the only factor that keeps working
Jegadeesh & Titman 1993 replicated across every market, every decade. Why the effect persists, how to construct it cleanly, and two production-grade templates on HK tech and CN blue chips.
If mean reversion is the anomaly that everyone learns first, momentum is the one that keeps working long after the textbooks have documented it to death. Jegadeesh & Titman published the original result in 1993; every replication since — including HK, CN, EM, crypto — has confirmed it.
“Momentum is the premier anomaly. No risk-based explanation has ever come close to accounting for it.”
Why it persists
Three mechanisms are load-bearing. Investor underreaction to news diffuses slowly, creating drift. Institutional rebalancing is sticky, so capital flows in the same direction for weeks. And behavioural bias — disposition effect, anchoring — systematically mis-prices recent winners and losers. None of these are going away.
The 12-1 construction
The canonical momentum signal is the past 12-month return minus the past 1-month return. Skipping the most recent month is not optional — without it, you pick up short-term reversion, which has the opposite sign and cancels your signal. This single detail is what separates working momentum from garbage.
From cross-sectional to time-series momentum
The 12-1 factor ranks stocks against each other. A separate family of strategies — time-series momentum — applies the same logic to a single asset over time: if the 20-day price is above the 60-day, stay long. Moving average crossovers are the simplest form of this, and remain the workhorse of most CTA funds.
Where momentum breaks
Momentum crashes. 2009, 2016, and 2020 each saw 6-12 month periods where the factor returned -20% or worse. The pattern is consistent: sharp reversals after extended one-way markets, when the "losers" portfolio contains beaten-down stocks that snap back on policy or liquidity shocks. Volatility-adjusted sizing is the standard defence — when realised vol of the momentum factor spikes, you reduce exposure.
Naive momentum runs hot for years then crashes hard. Every production momentum book uses some form of vol-scaling or regime filter. The templates above include two different approaches — study them, then decide which fits your risk tolerance.
Related tutorials
MA-Cross on HSTECH · regime-aware trend following
A disciplined 5/20 moving-average crossover on Hang Seng TECH, filtered by realised volatility. Designed to teach when trend-following works and when it quietly bleeds.
LightGBM Factor Stack · CSI 300
Gradient-boosted nonlinear factor interactions on A-shares, with proper walk-forward validation and turnover caps. A working production ML template.
50ETF Momentum Rotation · the most robust factor
Monthly cross-sectional momentum on CN blue chips with an Amihud illiquidity penalty. The simplest viable production momentum strategy.
Fork it into your workspace.
Every template in this article is one click away. Pick the one you want to play with and it lands in your workspace.