Momentum Time-Series
Time-series momentum: long when the return over the lookback is positive, out when it turns negative. Hurst, Ooi and Pedersen document it across 67 markets since 1880; it is also the only anomaly in the corpus that worked in 8 of the 10 worst crises of the last century. - The lookback is the whole strategy. Measured on XAUUSD H1 with 27.3 M real Dukascopy ticks (Jan-Jul 2016, spread 2.94 bps): a 250-bar lookback gave Sharpe 1.60 (CAGR 1.6 %, max drawdown −9.7 %, 84 trades) and 120 bars gave Sharpe 1.47 (max drawdown −8.3 %, 87 trades). - 🚨 It is not monotonic, and the research says so. A 60-bar lookback lost money (Sharpe −0.69) and a 20-bar one was mostly noise (Sharpe 0.72, CAGR −3.6 %). 250 and 120 work; 60 and 20 do not. Do not shorten it because it feels slow. - 🚨 One asset, not a basket. This is time-series momentum on the instrument you are looking at. The cross-sectional version — buying the coins that went up — has Sharpe 0.67 with 96.5 % of its return in 20 of 2,360 days, and its 2025-26 holdout is Sharpe −1.20. It is a regime strategy dressed as an edge. - Run it on daily or 4h. A 120-bar lookback is five months on the hourly chart and five days on the daily one. On five-minute bars it is noise. - The costs are Dukascopy's, which is tighter than a real broker: add 1-2 bps for yours before reading the numbers as yours. - The default stop (0.75 %) and target (1.5 %) are the ones the measurement used. The research's own next step was to widen the stop and narrow the target to 1:1 — it was not re-measured, so it is not the default here.