Trend Following in Gold and Silver Survives Testing — Filters Don't
A volatility-scaled Supertrend system returned +40R on gold 1H across 107 trades, while every entry filter tested failed to improve both metals in both periods.
By Nathan Brooks
4 min read

What's News
- Gold 1H system returned +40.0R across 107 trades (Jan 2025–Nov 2026) with 47% win rate and 8.6R max drawdown.
- Closing all trades at the 1R target cut gold 1H to +20.3R and turned gold 4H negative at −0.1R.
- Hold-and-add pyramiding produced the best results: +48.9R gold 1H, +34.1R silver 1H, +44.4R gold 4H.
- No entry filter improved results on both metals in both periods; the dollar-index filter cut gold 1H from +40.0R to +25.9R.
- On US equities, the 12-month trend filter lifted results from +89R to +153R across eight stocks and ETFs.
A rules-based trend-following system on gold returned +40.0R across 107 trades on the 1-hour chart from January 2025 to November 2026 — while every popular entry filter tested alongside it failed to improve results on both metals in both periods.
That's the headline finding from a TradingView analyst's rule-by-rule backtest, run on OANDA CFD data with trading costs deducted (0.40 per round trip on gold, 0.03 on silver). The system rests on three pillars: a stop that adapts to volatility, agreement across several timeframes, and risk defined in a fixed unit.
Why the structure works
The method uses a Supertrend built on Wilder's Average True Range (1978), averaged over 10 bars, with bands set at the bar midpoint ± 3 × ATR. Because distance is measured in ATR, the line widens on news days and tightens in quiet weeks. A flip on the 1H is accepted only if the Daily and 4H charts already agree — Elder's Triple Screen idea, where the higher frame sets direction and the lower frame times entry. Higher-frame values come from closed bars only, so nothing repaints.
Risk is defined in Van Tharp's R-multiples: a full stop loss is −1R, and targets sit at 1R, 2R and 3R. The academic grounding is time-series momentum: Moskowitz, Ooi and Pedersen (2012) documented it across futures markets, and Hurst, Ooi and Pedersen (2017) found trend following profitable in every decade across more than a century of data.
What does the trade management change?
The backtest's sharpest result: closing everything at the first target destroys the edge. Trend returns are positively skewed, and a few trades that run to the final target pay for many small losses.
- Hold to final target: Gold 1H +40.0R (max drawdown 8.6R); Silver 1H +24.1R (9.2); Gold 4H +28.3R (11.3); Silver 4H +15.3R (11.4)
- Close all at 1R: +20.3R / +3.4R / −0.1R / +2.9R — gold 4H turns negative
- Close at 1R, re-enter on pullback: +29.1R / +13.4R / +15.9R / +5.4R
- Hold + add on the 50% pullback: +48.9R / +34.1R / +44.4R / +17.9R
The add is disciplined pyramiding in the spirit of Livermore and the Turtles: after the first target, a second position of equal risk goes at the midpoint between entry and the best price reached. A plan and its add can lose up to 2R together, so size for it.
Also tested and rejected: a Chandelier trail after the final target, a target-free exit on the opposite signal, and a stop ladder past the final target. Trends that reached the final target gave back more afterwards than they added.
Win rates tell the asymmetry story. Gold 1H won just 47% of trades (silver 1H 42%, gold 4H 39%), yet all stayed profitable at +0.37R, +0.24R and +0.18R per trade. On gold 1H, 55% of trades touched the first target but only 18% reached the third.
Which filters failed?
Every filter — a 12-month trend requirement, skipping after shock candles above 2.5 ATR, avoiding high-volatility days (the Moreira & Muir idea), a dollar-index check, London–New York hours only, real yields below their 50-day average — helped in one period and hurt in another. The dollar-index filter cut gold 1H from +40.0R to +25.9R. The volatility filter crushed silver 1H to +3.9R.
The author's diagnosis: filters remove exactly the large, fast moves that begin in volatile, news-driven conditions, citing Bailey, Borwein, López de Prado and Zhu (2014) on backtest overfitting. US equities behaved differently — the 12-month trend filter lifted eight large stocks and index ETFs at 4H from +89R to +153R, with all eight positive, consistent with Jegadeesh and Titman (1993). A filter is good or bad for a given market, not in itself.
Silver's own settings, and the caveats
Silver needs wider targets: a 4R final target with the pullback add gave +55.2R (drawdown 12.4R) on 1H and +53.6R (15.1R) on 4H. The same tight stop doubled gold's drawdown. The setting came from 18 variants, so part of that edge may be luck. Position size follows from R, not leverage — leverage moves liquidation closer without changing the loss at the stop.
The limits are stated plainly: 1H samples cover roughly 22 months, mostly a rising gold market; data come from one broker's CFD prices; limit fills are assumed at their level. The author's conclusion for practitioners is direct — test every new rule on data it wasn't built on, and keep only what survives.
Source: tradingview.com
More from Nathan Brooks
Show full bio
News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.
2,426 articles