Gold trading strategies, tested on real data
Fourteen named gold strategies written as exact rules and tested on real XAUUSD prices with costs: scalps, session breakouts, ICT and price-action set-ups, and the martingale, grid and straddle systems that blow accounts. Losses and thin edges are shown as they are.
Results at a glance
Each row shows the test's key number (net US dollars per ounce after costs unless stated) and a one-line finding. Open any page for the exact rules, the full table, the split by period, when it fails, risk sizing and how to test it yourself.
| Strategy and data | Key result | What we found | Verdict |
|---|---|---|---|
| 1-minute EMA pullback scalp M1, 21 trading days (Oct 2026) | 343 trades: −$144 | Lost $0.12 a trade before costs; a $0.30 cost tripled the loss and $0.60 made it $247. On M1 the spread is most of the result. | Lost |
| 5-minute EMA 50/200 + RSI(7) scalp M5, 50 trading days + H1 2 years | 147 trades: −$60 | Negative before costs, halves disagreed, and the same rules lost $744 on two years of hourly data. | Lost |
| 15-minute session breakout M15, 50 trading days + H1 2 years | 119 trades: −$263 | Every version lost in both halves; the 200 EMA filter made it worse. A trailing exit, not a time stop, is what made hourly breakouts pay elsewhere. | Lost |
| Previous day high/low: break vs fade H1 2 years (513 days) + D1 since 2015 | Break: +$308; fade: −$665 | 87% of days took out yesterday's high or low. Trading the break was barely positive (profit factor 1.05, all in year 1); fading it lost. | Thin |
| New York opening range breakout M5, 49 sessions + H1 proxy 2 years | OR15 2R: −$114 | 15 and 30-minute ranges lost at 2R; 1R broke even. The 9:30 New York anchor is 13:30 GMT in US summer and 14:30 in winter. | Lost |
| Weekly open: bias vs retest H1 2 years (104 weeks) + D1 since 2015 | Retest: +$781 (53 trades) | As a bias it lost; as a retest level all six variants were positive (profit factor 1.24 to 2.71), but most profit came in 2026 and the sample is small. | Promising |
| Candle Range Theory (CRT) H4 2 years + D1 since 2015 | H4: −$1,817 | A mechanical version of the sweep-and-close-back candle lost on every timeframe and target; the stop was hit about twice as often as the far side. | Lost |
| ICT Judas swing (approximation) H1 2 years + M5 60 days | Raw: −$540; filtered: +$40 | Fading every London sweep of the Asian range lost; a midnight-open filter brought it to roughly break-even (profit factor 1.05). | Break-even |
| Turtle Soup D1 since 2015 + H1 2 years | H1: −$890; D1: −$218 | Fading failed 20-bar breakouts lost on both timeframes, and so did trading the breakout with a fixed 2R target. | Lost |
| Harmonic patterns H1, H4 2 years + D1 since 2015 | H1: 31 trades, +$84 | Detected automatically, the patterns were rare: 31 hourly trades, 6 on H4 (all lost), 8 daily. Far too few to judge either way. | Not enough data |
| Inside bar breakout D1 since 2015 + H4, H1 2 years | D1: +$1,061; D1 + EMA: +$1,290 | Daily inside bars were positive in both periods (best with a 50 EMA filter, mostly long trades); hourly inside bars lost $1,987. | Positive (daily) |
| Martingale (doubling) H1 2 years + D1 since 2015 | D1: ten years of gains lost in 2 weeks | Survived two years of hourly trades, then on daily data a 9-loss streak in February 2025 needed a 512-ounce trade. At 5 oz starting size, 61% of hourly windows ended in ruin. | Blows accounts |
| Grid and zone-recovery hedging H1 2 years | Every grid stopped out | A $10 buy grid banked 3,825 winners and $37,103, then was stopped out with 95 positions open; a $10/$20 zone-recovery hedge failed in 33 days. | Blows accounts |
| News straddle (NFP and CPI) 44 real releases, H1 + M5/M1 checks | $5 straddle: −$231 with slippage | Both sides were touched in 30 of 44 release hours; with realistic news slippage every version lost and losses ran up to 1.8x to 2.5x the planned risk. | Blows accounts |
For context, holding gold gained about $1,194 an ounce in the year to 30 October 2025 and about $14 in the following year, which included a run to about $5,627 in January 2026 and a fall back to around $4,000. Over the 5 and 15-minute windows (August to October 2026) gold fell by roughly $660 to $690 an ounce.
Pick a strategy
Every page has the exact rules, the test table, a plain-English reading, when it fails, position sizing and a link to the full lesson.
1-minute EMA pullback scalp
Gold 1-Minute Scalping Strategy: An EMA Pullback Scalp Tested, and What Costs Do to It
Read the test5-minute EMA 50/200 + RSI(7) scalp
Gold 5-Minute Scalping Strategy: The EMA 50/200 and RSI(7) Pullback, Tested on XAUUSD
Read the test15-minute session breakout
Gold 15-Minute Intraday Strategy: A Session Momentum Breakout Tested on XAUUSD
Read the testPrevious day high/low: break vs fade
Gold Previous Day High and Low Strategy: Break or Fade? Tested on XAUUSD
Read the testNew York opening range breakout
New York Opening Range Breakout on Gold: 15 and 30-Minute ORB Tested, With Time Zones Done Right
Read the testWeekly open: bias vs retest
Gold Weekly Open Strategy: Bias or Retest? Two Ways to Trade the XAUUSD Weekly Open, Tested
Read the testCandle Range Theory (CRT)
Candle Range Theory (CRT) on Gold: A Precise Definition and a Mechanical Test
Read the testICT Judas swing (approximation)
ICT Judas Swing on Gold: A Mechanical Version of the London False Move, Tested
Read the testTurtle Soup
Turtle Soup Strategy on Gold: Fading False 20-Day Breakouts, Tested on XAUUSD
Read the testHarmonic patterns
Harmonic Patterns on Gold: Gartley, Bat, Butterfly and Crab Explained, With an Honest Test
Read the testInside bar breakout
Gold Inside Bar Breakout Strategy: Daily, 4-Hour and Hourly Inside Bars Tested on XAUUSD
Read the testMartingale (doubling)
Gold Martingale Strategy: Why Doubling After Losses Blows Accounts, Simulated on Real XAUUSD Data
Read the testGrid and zone-recovery hedging
Gold Grid Trading and Hedging Strategies: Why They Blow Accounts, Simulated on Real XAUUSD Prices
Read the testNews straddle (NFP and CPI)
Gold News Straddle Strategy: Why Straddling NFP and CPI Blows Accounts, Tested on Real Releases
Read the testWhat the tests taught us about gold
Costs decide the fastest strategies. A $0.30 round trip is about 16% of the average 1-minute bar during London and New York hours, 6% of a 5-minute bar, 3.5% of a 15-minute bar and under 2% of an hourly bar. None of our intraday scalps had an edge even before costs, and costs made every one of them clearly negative.
Popular discretionary ideas lost as bare triggers. Candle Range Theory, the Judas swing, Turtle Soup and the previous-day fade all trade a sweep of an obvious level. In their mechanical form they lost, or at best broke even when a context filter (the New York midnight open) was added. The idea that stops get run is true; that every run reverses is not.
Levels worked better than biases, and slower charts better than fast ones. The weekly open as a retest level and daily inside bars were the two positive results. The same weekly open used as a directional bias lost, and the same inside bar on the hourly chart lost heavily.
Exits matter as much as entries. Breakouts with fixed targets or time stops lost here, while the hourly Donchian breakouts in our indicator settings tests were positive with a trailing exit that let winners run.
High win rates can hide ruin. The martingale, grid and zone-recovery simulations produced long runs of winning trades, and then lost the account. A strategy's worst losing streak and its worst open drawdown matter more than its win rate.
Limitations you should know
- Short intraday histories. Free 1-minute data covers about four weeks (21 trading days) and 5 and 15-minute data about 60 days (50 trading days). That is enough to show cost arithmetic and obvious failures, not enough to confirm a small edge.
- One market, one regime. Two years of hourly data were dominated by a powerful bull run and a sharp correction. Daily data since 2015 adds quieter years, and several results changed sign between periods.
- Approximations of discretionary ideas. CRT, the Judas swing and harmonic patterns were turned into strict rules. They cannot capture the judgement their teachers emphasise, and we say so on each page.
- Futures, not spot. Continuous futures include contract rolls and sit a few dollars from spot XAUUSD; broker day boundaries and candle alignment may differ from ours.
- Bar-level fills. Inside a bar we assume a standard price path and count the stop first when in doubt. Real fills, especially at news, can be worse; the straddle page shows how much that matters.
- Many variants. We report every variant we ran, including the losers. The more variants anyone tests, the more likely one looks good by chance.
Related lessons
The lessons below teach these methods in depth, with the discretion and context a mechanical test leaves out. The pages above are the stress tests.
- Gold scalping strategy (M1 and M5)
- New York session playbook
- London session gold trading
- Asian range breakout
- Liquidity sweeps on XAUUSD
- Power of 3 (AMD) and the Judas swing
- ICT concepts for XAUUSD
- Candlestick patterns on gold
- Breakout and retest
- Fibonacci on gold
- Gold news trading strategy
- XAUUSD risk management
- How to backtest a gold strategy
For indicator settings tested the same way (RSI, EMA, Ichimoku, Donchian and eight more), see gold indicator settings, tested. For the full strategy academy, start at XAUUSD trading strategies. Size every test trade with the position size calculator.
Practise these strategies with other traders
Our numbers are a starting point. The useful part is testing a strategy on your own data, in your own sessions, and having other traders challenge the results. Inside the XAUUSD Global community traders do exactly that every day. The first month is free, then $19.99 a month.
Frequently asked questions
Which gold trading strategy did best in your tests?
Two named strategies were positive over their full test: trading the break of a daily inside bar (+$1,061 per ounce over 321 trades since 2015, or +$1,290 with a 50 EMA filter) and trading a retest of the weekly open (+$781 per ounce over 53 trades in two years of hourly data). Both have caveats: the inside bar gains were mostly long trades in a bull market, and the weekly open sample is small and concentrated in 2026.
How were the strategies tested?
On COMEX gold futures (GC=F) from Yahoo Finance: 1-hour bars from 30 October 2024 to 30 October 2026, daily bars since 2015, and the short 1, 5 and 15-minute histories Yahoo provides (about 4 weeks of 1-minute and 60 days of 5 and 15-minute data). Each strategy was written as exact rules, signals used only information available at the time, orders filled at the next bar’s open or at the stop or limit price, and every trade paid $0.30 per ounce. Risk pages used a $10,000 account with 1:100 margin.
Why did so many named strategies lose?
Most popular strategies are taught as discretionary ideas. Reduced to a mechanical trigger, such as a candle that sweeps the previous candle’s high, or a sweep of the Asian range, they did not carry an edge after costs in our samples. Context, judgement and exits may add value, but they need to be proven with a journal, not assumed.
Why test martingale, grid and straddle systems at all?
Because they are common in gold robots and signal groups, and their trade histories look excellent until they fail. Simulating them on real prices shows the moment of failure: a 9-loss streak for martingale, a trend for grids, a whipsaw with slippage for news straddles.
Can I trade these results?
No. They are educational tests of simple rules on one data feed and limited history. They are not signals or recommendations. Use them as a starting point for your own testing, with your broker’s prices and costs, in a demo account first.