Time-Series Momentum (Trend) on Gold
Go long after positive trailing 12-month returns, short after negative — absolute/time-series momentum.
The most instructive result on the board — 8 of 11 gates passed, PF 2.41, yet the placebo proved the "edge" was pure long-gold beta, not momentum.
- Why it fails
- A random long/flat schedule at the same base rate beat the real signal. The 12-month rule added no timing value over simply being exposed to gold; it was a slightly-worse-than-random way to stay long a bull market.
- When / how it stopped
- Tested over 2017-01-02 → 2026-06-19 on Dukascopy daily gold (114 months). PF 2.41, Sharpe 0.89, −45.9% drawdown, 8 of 11 gates passed — but placebo FAIL.
Time-series momentum — also called absolute momentum or trend-following — is one of the most documented anomalies in finance: when an asset’s trailing 12-month return is positive you hold it long, when negative you go short or flat. The classic reference is Moskowitz, Ooi & Pedersen (2012), who found it across dozens of instruments and decades.
We ran a pre-registered, vol-targeted version on gold over 2017-01-02 → 2026-06-19 (Dukascopy daily, 114 monthly rebalances) through all 11 gates. On the surface it looked superb: PF 2.41, Sharpe 0.89, DSR 0.996, cost-robust, 70% of years positive, $5k → $25.3k. It passed 8 of 11 gates — more than any retired strategy we have.
Then it failed the kill-shot:
- The placebo randomises when you are long while holding the exposure level and frequency fixed. A random long/flat schedule at the same 70% base rate produced a PF whose 95th percentile (2.87) sat above the real strategy’s 2.41 — and 24% of random schedules beat it outright.
- Decomposition confirmed it: gold itself ran from ~$1,300 to ~$4,200, and 52% of all P&L came from the single 2025 surge. The −46% drawdown was gold’s own, not a risk control.
The 12-month filter was not selecting good months; it was a slightly-worse-than-random way to stay long a bull market. The lesson is the keeper here: PF 2.4 and Sharpe 0.89, and still no edge. The placebo was the only one of the eleven gates that could tell beta from alpha.
The full pre-registered report, gate scorecard and charts are in our validation write-up.
→ Read our full validation report: /strategy/tsmom-xauusd
Sources
- Our validation report — Time-Series Momentum (XAUUSD), 8/11 gates, placebo failed
- Moskowitz, Ooi & Pedersen (2012), 'Time Series Momentum', Journal of Financial Economics
Frequently asked
Does time-series momentum work in 2026?
Not as a standalone edge on gold. Tested through 11 pre-registered gates over 2017-01-02 → 2026-06-19, single-asset TSMOM passed 8 of 11 — profit factor 2.41, Sharpe 0.89, cost-robust — but failed the placebo. A random long/flat schedule at the same 70% base rate beat it. The apparent edge was levered long-gold beta, not momentum timing.
How can a strategy pass 8 gates and still have no edge?
Because most gates confirm the return stream is real — they cannot tell beta from alpha. Gold rose from roughly $1,300 to $4,200 over the sample; any rule that keeps you long about 70% of the time looks excellent on profit factor and Sharpe. Only the placebo gate holds the exposure fixed and randomises the timing — and it showed the 12-month signal added no timing value. That single gate is the one that caught it.
Not investment advice — your mileage may vary, but the burden of proof is on the person claiming an edge. This entry describes general research and published evidence (or its absence), not a recommendation. See the full disclaimer.