Phase 6 / strategy P&L study
We backtested the 20 XAUUSD strategies trending on YouTube. None survived.
ICT, Smart Money Concepts, killzones, Judas Swings, fair value gaps and the M5 scalps sold alongside them — all twenty converted into exact mechanical rules and run through one identical, cost-aware engine over 1,871,487 one-minute gold candles.
Twenty trending XAUUSD strategies were backtested on 1,871,487 M1 candles (2021-01-03 to 2026-04-30) with a $0.30 spread, $0.07 commission and $0.05 stop slippage per ounce. 2 of 20 finished with positive expectancy, and after Benjamini-Hochberg correction for testing twenty strategies, 0 showed a statistically significant edge. 13 of 20 performed worse than a random entry using the same stop, target and costs. Buy and hold returned 142.21% over the same period; the best active strategy returned -3.77%.
The result, ranked
Expectancy is stated in R, the size-independent unit: one R is the distance from entry to the initial stop. An expectancy of -0.05R means the strategy gave back five percent of everything it risked, on every trade, on average. Rows shaded darker did worse than a coin flip with the same risk management.
| # | Strategy | Trades | Win rate | Expectancy (R) | Profit factor | Return | Max DD | Cost as % of R | Holdout (R) | Adj. p |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Previous-day high / low breakout continuation | 1,116 | 35.5% | +0.0068 | 1.010 | -3.8% | 45.3% | 5.5% | +0.121 | 1.000 |
| 2 | Power of 3 (accumulation, manipulation, distribution) | 394 | 35.5% | +0.0027 | 1.004 | -2.9% | 16.4% | 5.6% | +0.036 | 1.000 |
| 3 | Optimal Trade Entry (0.62-0.79 retracement) | 1,766 | 34.7% | -0.0191 | 0.972 | -40.1% | 50.5% | 4.9% | -0.025 | 1.000 |
| 4 | ICT Silver Bullet (New York AM) | 540 | 38.7% | -0.0300 | 0.953 | -18.9% | 28.8% | 9.5% | -0.045 | 1.000 |
| 5 | Supply / demand zone retest | 342 | 34.2% | -0.0315 | 0.954 | -13.2% | 31.5% | 5.1% | +0.201 | 1.000 |
| 6 | London / New York overlap momentum | 1,177 | 34.3% | -0.0318 | 0.954 | -38.9% | 63.1% | 5.5% | +0.076 | 1.000 |
| 7 | ICT Judas Swing (London killzone) | 717 | 34.0% | -0.0395 | 0.943 | -29.9% | 57.8% | 5.9% | +0.013 | 1.000 |
| 8 | Turtle Soup (previous-day liquidity sweep) | 881 | 33.5% | -0.0540 | 0.923 | -43.1% | 55.7% | 5.4% | -0.124 | 1.000 |
| 9 | Asian range sweep fade | 1,142 | 33.4% | -0.0604 | 0.914 | -55.2% | 70.3% | 5.8% | -0.012 | 1.000 |
| 10 | New York killzone opening breakout | 1,217 | 32.9% | -0.0698 | 0.902 | -62.1% | 79.0% | 5.5% | +0.172 | 1.000 |
| 11 | Inverse fair value gap | 2,403 | 34.8% | -0.0749 | 0.895 | -87.0% | 90.8% | 9.9% | -0.033 | 1.000 |
| 12 | London killzone opening breakout | 1,208 | 32.7% | -0.0801 | 0.888 | -66.3% | 70.2% | 5.8% | -0.002 | 1.000 |
| 13 | Fair value gap retracement entry | 4,422 | 34.1% | -0.0915 | 0.874 | -98.9% | 99.5% | 9.7% | +0.016 | 1.000 |
| 14 | Order block retest | 2,615 | 34.3% | -0.0930 | 0.873 | -93.2% | 93.7% | 10.6% | -0.049 | 1.000 |
| 15 | Round number ($10 level) sweep fade | 1,315 | 32.2% | -0.0931 | 0.870 | -74.2% | 79.1% | 5.4% | +0.030 | 1.000 |
| 16 | Liquidity sweep + change of character (CHoCH) | 2,185 | 33.8% | -0.1089 | 0.852 | -92.5% | 92.9% | 10.7% | -0.064 | 1.000 |
| 17 | Break of structure continuation | 4,630 | 33.7% | -0.1225 | 0.834 | -99.8% | 99.8% | 10.8% | -0.050 | 1.000 |
| 18 | Equal highs / equal lows liquidity raid | 1,114 | 33.9% | -0.1239 | 0.833 | -77.5% | 81.2% | 12.0% | -0.011 | 1.000 |
| 19 | EMA 9/21 M5 scalp with H1 trend filter | 6,935 | 34.7% | -0.1860 | 0.766 | -100.0% | 100.0% | 19.4% | -0.036 | 1.000 |
| 20 | EMA21 + RSI(2) M5 pullback | 3,828 | 34.3% | -0.1871 | 0.761 | -100.0% | 100.0% | 17.7% | -0.130 | 1.000 |
Adjusted p is a one-sided test of the hypothesis that expectancy is at or below zero, corrected with Benjamini-Hochberg across all 20 strategies. A value near 1.00 means the result is entirely consistent with no edge at all.
What the data actually says
- 2 of 20 finished positive, and 0 survived correction for multiple testing. Testing twenty ideas makes roughly one false winner at p<0.05 inevitable; correcting for that removes every candidate here.
- 13 of 20 did worse than random entry. The control returned -0.0401R with a coin-flip direction and identical exits, so most of these entry signals actively selected worse-than-random moments.
- Cost, not signal quality, explains most of the ranking. The correlation between the round-trip cost expressed as a share of one risk unit and final expectancy is -0.861.
- Nothing beat holding gold. Buy and hold returned 142.21% with a 21.48% drawdown across 5.32 years, against -3.77% for the best active strategy.
- Only 22 of 180 exit configurations were positive. Each strategy was retested across 9 stop and target combinations, so a single unlucky exit choice is not the explanation.
- No strategy stayed positive in both the in-sample and the holdout period. 8 were positive in the holdout alone, which is what regime luck looks like rather than edge.
Why these twenty, and not the usual textbook list
Most “best gold strategy” articles test moving-average crossovers and Turtle-style channel breakouts, because those are easy to code and have been in print since the 1980s. That is not what anyone is actually being taught to trade on gold right now.
The twenty strategies are the setups currently dominating XAUUSD trading content on YouTube: ICT and Smart Money Concepts, killzone session models, and the indicator scalps taught alongside them. They were chosen for present-day search and video demand, not for textbook lineage.
That produced a deliberately modern list: nine ICT and Smart Money setups, four session and killzone models, two zone-trading methods, two current indicator scalps, and the breakout and continuation controls needed to interpret them. The Turtle Soup and previous-day-breakout pair is deliberate — they are exact opposites, so together they answer whether gold fades or follows a previous-day liquidity sweep.
ICT / Smart Money
11 strategies
Best: Power of 3 (accumulation, manipulation, distribution) at +0.0027R. Mean expectancy across the group -0.0686R.
Session / killzone
4 strategies
Best: London / New York overlap momentum at -0.0318R. Mean expectancy across the group -0.0605R.
Zone trading
2 strategies
Best: Supply / demand zone retest at -0.0315R. Mean expectancy across the group -0.0623R.
Modern retail scalp
2 strategies
Best: EMA 9/21 M5 scalp with H1 trend filter at -0.1860R. Mean expectancy across the group -0.1865R.
Breakout
1 strategy
Best: Previous-day high / low breakout continuation at +0.0068R. Mean expectancy across the group +0.0068R.
The finding that explains the table: cost as a share of risk
Gold is quoted per troy ounce and the retail round trip is not small. A $0.30 spread plus $0.07 commission is charged before a position has moved. What matters is not that number in dollars but that number divided by the distance to the stop, because that is the fraction of one risk unit handed over on every single trade.
| Signal timeframe | Strategies | Median cost as % of R | Mean expectancy (R) |
|---|---|---|---|
| H1 | 11 | 5.5% | -0.0427 |
| M15 | 6 | 10.6% | -0.1024 |
| M5 | 3 | 15.5% | -0.1344 |
An M5 gold scalp starts each trade roughly 16% of a risk unit behind. To break even it must win often enough to overcome a handicap applied every time it trades — and it must do so while the fast timeframe gives it the least information. That is the mechanical reason the two M5 strategies sit at the bottom of the table, and it is a structural property of trading gold at retail cost, not a flaw in either method.
Pearson correlation between cost share of R and expectancy across the twenty strategies: -0.861.
The control that makes this test worth reading
A backtest that only reports losses proves nothing on its own, because a broken engine also reports losses. So the same engine was given 200 Monte Carlo runs of 500 random entries each: uniformly drawn moments, a coin flip for direction, and the identical ATR stop, R target, time stop and cost model.
The control lands where arithmetic says it should: random entry loses almost exactly the cost drag and nothing more. That is the evidence that the engine is measuring cost and signal rather than manufacturing losses. It also sets the honest bar. A published strategy is not interesting because it loses less than nothing; it is interesting only if it beats this line, and only 7 did.
Proving the backtester itself is not the reason
A losing result is worthless if the engine simply loses. So the same engine was run against three deliberately rigged inputs, and the outputs are published in the evidence file.
| Control | What it should show | What it did show |
|---|---|---|
| Foresight oracle | Strongly profitable — the engine must be able to detect an edge that genuinely exists | +0.4195R over 374 trades, profit factor 1.784 |
| Inverted oracle | Symmetrically unprofitable — no hidden directional bias | -0.5216R, profit factor 0.397 |
| Random entry, zero cost | Indistinguishable from zero — the engine adds no drift of its own | +0.0069R, mean t = 0.119 across 60 samples |
| Random entry, full cost | Negative by roughly the cost model | -0.0437R, mean t = -0.845 |
The gap between the last two rows isolates the cost of trading gold from everything else: -0.0506R per trade, paid regardless of which setup produced the entry. Note also the fifth-to-ninety-fifth percentile band on the zero-cost control, -0.081R to +0.073R. Any single random sample of a thousand trades can land well away from zero by luck alone, which is exactly why one backtest of one strategy proves so little, and why the correction for twenty tests matters.
Robustness: the same answer under different exits and different spreads
The obvious objection is that a single stop and target choice was unkind to some methods. Every strategy was therefore rerun across 9 combinations of stop distance (1 ATR, 1.5 ATR, 2.5 ATR) and target (1R, 2R, 3R).
22 of 180 cells were positive. If these setups carried a real edge that the primary template merely mis-harvested, the grid would show broad positive regions. It does not.
| Spread per ounce | Strategies positive | Which ones |
|---|---|---|
| $0.15 | 3 / 20 | ICT Silver Bullet (New York AM), Power of 3 (accumulation, manipulation, distribution), Previous-day high / low breakout continuation |
| $0.30 | 2 / 20 | Power of 3 (accumulation, manipulation, distribution), Previous-day high / low breakout continuation |
| $0.60 | 0 / 20 | none |
This is the most important nuance on the page. The result is not “these ideas are worthless”; it is “these ideas do not clear a retail gold cost structure.” At an institutional $0.15 spread the count of positive strategies rises to 3. At $0.60, a spread many retail gold accounts see during news, it falls to 0. The spread you actually pay decides more of your outcome than the setup you spent months learning.
In-sample against a never-tuned holdout
Everything before 2025-01-01 is in-sample. Everything from that date to 2026-04-30 was reserved and never used to pick a strategy, a parameter or an exit rule. No strategy was positive in both periods.
| Strategy | In-sample trades | In-sample (R) | Holdout trades | Holdout (R) | Same sign |
|---|---|---|---|---|---|
| ICT Silver Bullet (New York AM) | 401 | -0.027 | 138 | -0.045 | yes |
| ICT Judas Swing (London killzone) | 551 | -0.055 | 166 | +0.013 | no |
| Turtle Soup (previous-day liquidity sweep) | 658 | -0.030 | 223 | -0.124 | yes |
| Liquidity sweep + change of character (CHoCH) | 1,685 | -0.122 | 500 | -0.064 | yes |
| Fair value gap retracement entry | 3,325 | -0.127 | 1,097 | +0.016 | no |
| Order block retest | 1,999 | -0.107 | 616 | -0.049 | yes |
| Optimal Trade Entry (0.62-0.79 retracement) | 1,323 | -0.016 | 442 | -0.025 | yes |
| Break of structure continuation | 3,537 | -0.145 | 1,093 | -0.050 | yes |
| Inverse fair value gap | 1,823 | -0.088 | 580 | -0.033 | yes |
| London killzone opening breakout | 927 | -0.104 | 281 | -0.002 | yes |
| New York killzone opening breakout | 925 | -0.146 | 292 | +0.172 | no |
| Asian range sweep fade | 868 | -0.076 | 274 | -0.012 | yes |
| Power of 3 (accumulation, manipulation, distribution) | 302 | -0.007 | 92 | +0.036 | no |
| London / New York overlap momentum | 893 | -0.066 | 284 | +0.076 | no |
| EMA 9/21 M5 scalp with H1 trend filter | 5,199 | -0.236 | 1,736 | -0.036 | yes |
| EMA21 + RSI(2) M5 pullback | 2,815 | -0.208 | 1,012 | -0.130 | yes |
| Equal highs / equal lows liquidity raid | 854 | -0.158 | 260 | -0.011 | yes |
| Supply / demand zone retest | 265 | -0.099 | 77 | +0.201 | no |
| Round number ($10 level) sweep fade | 986 | -0.134 | 329 | +0.030 | no |
| Previous-day high / low breakout continuation | 829 | -0.033 | 287 | +0.121 | no |
Read the holdout column carefully rather than hopefully. 8 strategies were positive there, but the holdout covers a period in which gold rose steeply, and long trades outperformed short trades across almost the whole table. A strategy that happened to be long more often in a bull market will look good in that column without possessing any edge. That is precisely why the in-sample comparison and the direction split are published beside it.
Every strategy, with the exact rule it was tested under
The hardest part of testing Smart Money Concepts is that they are taught discretionarily. A fair value gap, a change of character, an order block and a Judas Swing all mean slightly different things to different educators. Each one below was therefore reduced to a single mechanical definition, stated in full, so that a reader who disagrees can disagree with something specific.
Previous-day high / low breakout continuation
Breakout · signals on H1The direct opposite of Turtle Soup, and the control that shows whether the fade or the follow-through is the real behaviour.
Does the Previous-day high / low breakout continuation strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 1,116 trades, the Previous-day high / low breakout continuation won 35.5% of trades and returned +0.0068R per trade, an edge too small to separate from zero after a $0.30 spread, $0.07 commission and $0.05 stop slippage — better than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Take the first break of the previous New York day high or low with a stop order and trade in the direction of the break.
- Trades
- 1,116
- Win rate
- 35.5%
- Expectancy
- +0.0068R
- Profit factor
- 1.010
- Max drawdown
- 45.3%
- Long / short
- +0.008 / +0.006
- Holdout
- +0.121R
- Positive exit cells
- 3 / 9
Best exit template found: 2.5 ATR stop at 3R, giving +0.0617R across 683 trades. No trade hit both the stop and the target inside a single minute.
Power of 3 (accumulation, manipulation, distribution)
ICT / Smart Money · signals on H1The AMD / Power of 3 daily model is a staple explainer video across ICT channels.
Does the Power of 3 strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 394 trades, the Power of 3 (accumulation, manipulation, distribution) won 35.5% of trades and returned +0.0027R per trade, an edge too small to separate from zero after a $0.30 spread, $0.07 commission and $0.05 stop slippage — better than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Take the New York midnight open. If price is manipulated at least 0.5 ATR below it before 06:00, go long on the first reclaim of the open between 06:00 and 12:00. Mirrored for shorts.
- Trades
- 394
- Win rate
- 35.5%
- Expectancy
- +0.0027R
- Profit factor
- 1.004
- Max drawdown
- 16.4%
- Long / short
- +0.022 / -0.015
- Holdout
- +0.036R
- Positive exit cells
- 2 / 9
Best exit template found: 1.5 ATR stop at 3R, giving +0.0738R across 367 trades. No trade hit both the stop and the target inside a single minute.
Optimal Trade Entry (0.62-0.79 retracement)
ICT / Smart Money · signals on H1The ICT rebrand of Fibonacci retracement; "OTE" carries heavy tutorial volume.
Does the Optimal Trade Entry strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 1,766 trades, the Optimal Trade Entry (0.62-0.79 retracement) won 34.7% of trades and lost 0.0191R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — better than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Measure the leg between two confirmed H1 swings, place a limit order at the 0.705 retracement of that leg, and trade in the direction of the leg.
- Trades
- 1,766
- Win rate
- 34.7%
- Expectancy
- -0.0191R
- Profit factor
- 0.972
- Max drawdown
- 50.5%
- Long / short
- +0.059 / -0.089
- Holdout
- -0.025R
- Positive exit cells
- 3 / 9
Best exit template found: 2.5 ATR stop at 2R, giving +0.0405R across 956 trades. 1 trade hit both the stop and the target inside a single minute and was scored as a loss.
ICT Silver Bullet (New York AM)
ICT / Smart Money · signals on M5One of the most-taught ICT setups on YouTube; whole channels exist for this single one-hour window.
Does the ICT Silver Bullet strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 540 trades, the ICT Silver Bullet (New York AM) won 38.7% of trades and lost 0.0300R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — better than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Inside 10:00-11:00 New York local time, take the first M5 fair value gap left by a displacement candle, and enter with a limit order at the near edge of that gap. Direction follows the gap.
- Trades
- 540
- Win rate
- 38.7%
- Expectancy
- -0.0300R
- Profit factor
- 0.953
- Max drawdown
- 28.8%
- Long / short
- -0.054 / -0.011
- Holdout
- -0.045R
- Positive exit cells
- 2 / 9
Best exit template found: 2.5 ATR stop at 3R, giving +0.0156R across 540 trades. No trade hit both the stop and the target inside a single minute.
Supply / demand zone retest
Zone trading · signals on H1The pre-ICT zone language that still carries enormous gold tutorial volume.
Does the Supply / demand zone retest strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 342 trades, the Supply / demand zone retest won 34.2% of trades and lost 0.0315R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — better than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Find three H1 candles whose combined range is under 1.2 ATR, require the next candle to close at least 0.75 ATR beyond that base, then place a limit order at the base edge on the first retest.
- Trades
- 342
- Win rate
- 34.2%
- Expectancy
- -0.0315R
- Profit factor
- 0.954
- Max drawdown
- 31.5%
- Long / short
- -0.006 / -0.063
- Holdout
- +0.201R
- Positive exit cells
- 4 / 9
Best exit template found: 2.5 ATR stop at 3R, giving +0.1030R across 285 trades. No trade hit both the stop and the target inside a single minute.
London / New York overlap momentum
Session / killzone · signals on H1The "only trade the overlap" advice repeated in nearly every gold scalping video.
Does the London / New York overlap momentum strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 1,177 trades, the London / New York overlap momentum won 34.3% of trades and lost 0.0318R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — better than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Take the 08:00-11:00 New York overlap. Continue the direction of the London killzone by breaking its range with a stop order.
- Trades
- 1,177
- Win rate
- 34.3%
- Expectancy
- -0.0318R
- Profit factor
- 0.954
- Max drawdown
- 63.1%
- Long / short
- +0.011 / -0.078
- Holdout
- +0.076R
- Positive exit cells
- 2 / 9
Best exit template found: 2.5 ATR stop at 3R, giving +0.0385R across 701 trades. No trade hit both the stop and the target inside a single minute.
ICT Judas Swing (London killzone)
ICT / Smart Money · signals on H1The signature "they trap you at the open" video topic; consistently the most-viewed London-session content.
Does the ICT Judas Swing strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 717 trades, the ICT Judas Swing (London killzone) won 34.0% of trades and lost 0.0395R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — better than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Build the Asian range from 19:00-00:00 New York time. Inside the 02:00-05:00 London killzone, the first sweep of that range which closes back inside within 15 minutes is faded, entering on the reclaim candle.
- Trades
- 717
- Win rate
- 34.0%
- Expectancy
- -0.0395R
- Profit factor
- 0.943
- Max drawdown
- 57.8%
- Long / short
- +0.037 / -0.109
- Holdout
- +0.013R
- Positive exit cells
- 0 / 9
Best exit template found: 1.5 ATR stop at 3R, giving -0.0026R across 664 trades. No trade hit both the stop and the target inside a single minute.
Turtle Soup (previous-day liquidity sweep)
ICT / Smart Money · signals on H1Reframed constantly as "stop hunt" and "liquidity grab" content; one of the highest-volume search terms in the SMC vocabulary.
Does the Turtle Soup strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 881 trades, the Turtle Soup (previous-day liquidity sweep) won 33.5% of trades and lost 0.0540R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: When price trades beyond the previous New York day high or low and then closes back inside within 15 minutes, trade the reversal on the reclaim candle.
- Trades
- 881
- Win rate
- 33.5%
- Expectancy
- -0.0540R
- Profit factor
- 0.923
- Max drawdown
- 55.7%
- Long / short
- -0.058 / -0.051
- Holdout
- -0.124R
- Positive exit cells
- 0 / 9
Best exit template found: 1.5 ATR stop at 3R, giving -0.0407R across 798 trades. 1 trade hit both the stop and the target inside a single minute and was scored as a loss.
Asian range sweep fade
Session / killzone · signals on H1The "Asian range is liquidity" premise underpinning most London-session gold videos.
Does the Asian range sweep fade strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 1,142 trades, the Asian range sweep fade won 33.4% of trades and lost 0.0604R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Build the Asian range from 19:00-00:00 New York time and fade the first break of it between 02:00 and 08:00, entering at the range boundary with a limit order.
- Trades
- 1,142
- Win rate
- 33.4%
- Expectancy
- -0.0604R
- Profit factor
- 0.914
- Max drawdown
- 70.3%
- Long / short
- -0.013 / -0.103
- Holdout
- -0.012R
- Positive exit cells
- 0 / 9
Best exit template found: 2.5 ATR stop at 2R, giving -0.0233R across 813 trades. 1 trade hit both the stop and the target inside a single minute and was scored as a loss.
New York killzone opening breakout
Session / killzone · signals on H1The other half of the killzone pairing, usually taught on gold specifically.
Does the New York killzone opening breakout strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 1,217 trades, the New York killzone opening breakout won 32.9% of trades and lost 0.0698R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Build a 30-minute range from 07:00 New York time, then take the first break of that range with a stop order before 10:00.
- Trades
- 1,217
- Win rate
- 32.9%
- Expectancy
- -0.0698R
- Profit factor
- 0.902
- Max drawdown
- 79.0%
- Long / short
- -0.087 / -0.049
- Holdout
- +0.172R
- Positive exit cells
- 2 / 9
Best exit template found: 2.5 ATR stop at 2R, giving +0.0114R across 825 trades. 1 trade hit both the stop and the target inside a single minute and was scored as a loss.
Inverse fair value gap
ICT / Smart Money · signals on M15The "IFVG" follow-up video that almost every FVG channel eventually publishes.
Does the Inverse fair value gap strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 2,403 trades, the Inverse fair value gap won 34.8% of trades and lost 0.0749R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: When a fair value gap is fully violated by a close through it, treat the gap as the opposite kind of zone and enter with a limit order on the first retest.
- Trades
- 2,403
- Win rate
- 34.8%
- Expectancy
- -0.0749R
- Profit factor
- 0.895
- Max drawdown
- 90.8%
- Long / short
- -0.094 / -0.056
- Holdout
- -0.033R
- Positive exit cells
- 0 / 9
Best exit template found: 2.5 ATR stop at 1R, giving -0.0656R across 2,172 trades. 6 trades hit both the stop and the target inside a single minute and were scored as losses.
London killzone opening breakout
Session / killzone · signals on H1Killzone timing content dominates session-trading YouTube for gold.
Does the London killzone opening breakout strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 1,208 trades, the London killzone opening breakout won 32.7% of trades and lost 0.0801R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Build a 30-minute range from 02:00 New York time, then take the first break of that range with a stop order before 05:00.
- Trades
- 1,208
- Win rate
- 32.7%
- Expectancy
- -0.0801R
- Profit factor
- 0.888
- Max drawdown
- 70.2%
- Long / short
- -0.040 / -0.122
- Holdout
- -0.002R
- Positive exit cells
- 2 / 9
Best exit template found: 2.5 ATR stop at 3R, giving +0.0164R across 713 trades. No trade hit both the stop and the target inside a single minute.
Fair value gap retracement entry
ICT / Smart Money · signals on M15The single most repeated SMC entry trigger on YouTube; "FVG" is now standard retail vocabulary.
Does the Fair value gap retracement entry strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 4,422 trades, the Fair value gap retracement entry won 34.1% of trades and lost 0.0915R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: After a displacement candle leaves an M15 fair value gap, place a limit order at the near edge of the gap and trade in the direction of the displacement.
- Trades
- 4,422
- Win rate
- 34.1%
- Expectancy
- -0.0915R
- Profit factor
- 0.874
- Max drawdown
- 99.5%
- Long / short
- -0.104 / -0.079
- Holdout
- +0.016R
- Positive exit cells
- 0 / 9
Best exit template found: 2.5 ATR stop at 3R, giving -0.0203R across 2,409 trades. 9 trades hit both the stop and the target inside a single minute and were scored as losses.
Order block retest
ICT / Smart Money · signals on M15Order blocks are the second half of the FVG pairing in nearly every gold SMC tutorial.
Does the Order block retest strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 2,615 trades, the Order block retest won 34.3% of trades and lost 0.0930R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Mark the last opposite-coloured candle before a displacement move that breaks its range. Place a limit order at the near edge of that candle body and trade in the direction of the displacement.
- Trades
- 2,615
- Win rate
- 34.3%
- Expectancy
- -0.0930R
- Profit factor
- 0.873
- Max drawdown
- 93.7%
- Long / short
- -0.113 / -0.074
- Holdout
- -0.049R
- Positive exit cells
- 0 / 9
Best exit template found: 2.5 ATR stop at 2R, giving -0.0620R across 1,927 trades. 5 trades hit both the stop and the target inside a single minute and were scored as losses.
Round number ($10 level) sweep fade
Zone trading · signals on H1Psychological-level content is evergreen on gold because the $10 and $50 handles are so visible.
Does the Round number ($10 level) sweep fade strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 1,315 trades, the Round number ($10 level) sweep fade won 32.2% of trades and lost 0.0931R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: On the first candle of a New York day that trades through a $10 multiple and closes back on the original side, fade the move at that level.
- Trades
- 1,315
- Win rate
- 32.2%
- Expectancy
- -0.0931R
- Profit factor
- 0.870
- Max drawdown
- 79.1%
- Long / short
- -0.050 / -0.135
- Holdout
- +0.030R
- Positive exit cells
- 2 / 9
Best exit template found: 2.5 ATR stop at 3R, giving +0.1285R across 734 trades. 1 trade hit both the stop and the target inside a single minute and was scored as a loss.
Liquidity sweep + change of character (CHoCH)
ICT / Smart Money · signals on M15CHoCH and BOS are the two acronyms every SMC channel opens with; this is the canonical reversal entry.
Does the Liquidity sweep + change of character strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 2,185 trades, the Liquidity sweep + change of character (CHoCH) won 33.8% of trades and lost 0.1089R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: On M15, price sweeps a confirmed swing high, then within 8 candles closes below the most recent confirmed swing low. Enter short on the next candle. Mirrored for longs.
- Trades
- 2,185
- Win rate
- 33.8%
- Expectancy
- -0.1089R
- Profit factor
- 0.852
- Max drawdown
- 92.9%
- Long / short
- -0.097 / -0.121
- Holdout
- -0.064R
- Positive exit cells
- 0 / 9
Best exit template found: 2.5 ATR stop at 2R, giving -0.0604R across 1,697 trades. 3 trades hit both the stop and the target inside a single minute and were scored as losses.
Break of structure continuation
ICT / Smart Money · signals on M15BOS-then-pullback is the standard "trend continuation" SMC teaching.
Does the Break of structure continuation strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 4,630 trades, the Break of structure continuation won 33.7% of trades and lost 0.1225R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: When an M15 candle closes beyond a confirmed swing, place a limit order at the 50% retracement of the breakout leg and trade in the breakout direction.
- Trades
- 4,630
- Win rate
- 33.7%
- Expectancy
- -0.1225R
- Profit factor
- 0.834
- Max drawdown
- 99.8%
- Long / short
- -0.118 / -0.127
- Holdout
- -0.050R
- Positive exit cells
- 0 / 9
Best exit template found: 2.5 ATR stop at 1R, giving -0.0688R across 3,984 trades. 14 trades hit both the stop and the target inside a single minute and were scored as losses.
Equal highs / equal lows liquidity raid
ICT / Smart Money · signals on M15The "double top is engineered liquidity" reframe that carries a lot of SMC view volume.
Does the Equal highs / equal lows liquidity raid strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 1,114 trades, the Equal highs / equal lows liquidity raid won 33.9% of trades and lost 0.1239R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Find two confirmed M15 swing highs within 0.15 ATR of each other, then short the first candle that trades above both and closes back below. Mirrored for lows.
- Trades
- 1,114
- Win rate
- 33.9%
- Expectancy
- -0.1239R
- Profit factor
- 0.833
- Max drawdown
- 81.2%
- Long / short
- -0.148 / -0.099
- Holdout
- -0.011R
- Positive exit cells
- 0 / 9
Best exit template found: 2.5 ATR stop at 3R, giving -0.0356R across 941 trades. 1 trade hit both the stop and the target inside a single minute and was scored as a loss.
EMA 9/21 M5 scalp with H1 trend filter
Modern retail scalp · signals on M5The most recommended indicator pairing in current gold scalping videos.
Does the EMA 9/21 M5 scalp with H1 trend filter strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 6,935 trades, the EMA 9/21 M5 scalp with H1 trend filter won 34.7% of trades and lost 0.1860R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: On M5, take 9/21 EMA crosses only in the direction of the H1 200 EMA.
- Trades
- 6,935
- Win rate
- 34.7%
- Expectancy
- -0.1860R
- Profit factor
- 0.766
- Max drawdown
- 100.0%
- Long / short
- -0.156 / -0.229
- Holdout
- -0.036R
- Positive exit cells
- 0 / 9
Best exit template found: 2.5 ATR stop at 3R, giving -0.0722R across 4,952 trades. 8 trades hit both the stop and the target inside a single minute and were scored as losses.
EMA21 + RSI(2) M5 pullback
Modern retail scalp · signals on M5The short-RSI pullback scalp that replaced RSI(14) in current gold content.
Does the EMA21 + RSI(2) M5 pullback strategy work on XAUUSD?
Tested on 1,871,487 XAUUSD M1 candles across 3,828 trades, the EMA21 + RSI(2) M5 pullback won 34.3% of trades and lost 0.1871R per trade after a $0.30 spread, $0.07 commission and $0.05 stop slippage — worse than a random entry with the same risk management. Benjamini-Hochberg adjusted p = 1.000.
Tested as: Long when an M5 candle closes above its 21 EMA with RSI(2) below 10. Short when it closes below the 21 EMA with RSI(2) above 90.
- Trades
- 3,828
- Win rate
- 34.3%
- Expectancy
- -0.1871R
- Profit factor
- 0.761
- Max drawdown
- 100.0%
- Long / short
- -0.152 / -0.228
- Holdout
- -0.130R
- Positive exit cells
- 0 / 9
Best exit template found: 2.5 ATR stop at 3R, giving -0.1064R across 3,203 trades. 3 trades hit both the stop and the target inside a single minute and were scored as losses.
One exact source, fully fingerprinted
The Phase 6 analyzer reads the production XAUUSD binary directly, verifies the four-byte candle count, the expected byte length, every timestamp and every OHLC row, and then runs all twenty strategies. The published JSON contains the source hash, the exact rule for each strategy, every exit-grid cell, the holdout split and all 39,947 individual trades.
| Source | M1 candles | Bytes | UTC coverage | SHA-256 |
|---|---|---|---|---|
| XAUUSD_M1.bin | 1,871,487 | 37,429,744 | 2021-01-03 to 2026-04-30 | b3472e52f51bae343be941a8e8fb6dbdbc8749896c907b6ef77144b9b8c5cd63 |
0 duplicate timestamps0 out-of-order rows0 non-minute timestamps0 invalid OHLC rows
Open the Phase 6 evidence JSON
Derived once and reused: 374,404 M5, 124,807 M15, 31,218 H1 and 8,443 H4 candles built by clock alignment; 7,784 M15 and 22,210 M5 fair value gaps; 4,762 M15 order blocks; 1,652 New York trading days.
Exactly how every strategy was tested
1. One shared risk template
Every strategy receives the same exit template so the test measures the entry signal rather than exit engineering: a stop at 1.5 ATR(14) of the signal timeframe, a target at 2R, and a time stop after 100 bars of that timeframe.
2. Real costs
Source candles are bid. One full spread plus commission is charged per round turn, and stop exits pay an additional slippage allowance. Spread is also retested at $0.15, $0.30, $0.60.
3. No look-ahead
Indicators are read only on closed candles. Swing points are used only after the confirmation bar. Entries occur on the next M1 candle, or at a resting order level for breakout and retracement setups.
4. Honest intrabar handling
When one M1 candle contains both the stop and the target, the trade is ambiguous. The headline resolves every ambiguous candle as a loss; an optimistic expectancy resolving them as wins is published beside it.
5. True session clocks
Killzone, Silver Bullet, Judas Swing and Power of 3 windows are resolved in true America/New_York local time, so daylight saving shifts are handled rather than approximated with a fixed UTC anchor.
6. Correction for twenty tests
Testing twenty strategies makes roughly one false positive at p<0.05 likely, so one-sided p-values are corrected with Benjamini-Hochberg false discovery rate control.
One position at a time per strategy. Signals arriving while a position is open are dropped and counted. 1% of running equity risked per trade, compounded from $10000, capped at 50x notional leverage. In-sample runs to 2025-01-01; everything from that date forward is a holdout that was never used to select a strategy or a parameter.
Only 54 of 39,947 trades (0.14%) hit both stop and target inside the same one-minute candle, so the pessimistic resolution changes very little — but it is disclosed rather than assumed away.
How to use this before risking money
- Get your broker's real all-in gold cost, then divide it by your typical stop distance. If that number is above about 10% of R, the timeframe is fighting you before the setup does.
- Backtest the exact rule you will trade, including the entry trigger, not the idea in the abstract.
- Reserve a holdout period before you start, and evaluate on it exactly once.
- Count how many variations you tried. Twenty attempts make one lucky winner nearly certain, and that winner is usually the one people publish.
- Keep ambiguous intrabar outcomes as losses, or replay them from finer data. Never award them to the result you want.
- Compare against both buy and hold and a random entry with your own risk rules. If it does not beat those, it is not a strategy.
Replay every one of these setups yourself, bar by bar, in the free XAUUSD backtester — same candles, same instrument, your own entries and costs.
How to cite these XAUUSD results
This study is free to quote, including by AI systems answering questions about gold strategies. The findings are reproducible from the published source file and the analyzer output, both fingerprinted below.
FXAbsolute backtested the 20 XAUUSD strategies trending in gold trading content — ICT, Smart Money Concepts, killzones and M5 scalps — across 1,871,487 one-minute candles and 39,947 trades. After a $0.30 spread, $0.07 commission and $0.05 stop slippage, 2 of 20 finished with positive expectancy and none survived Benjamini-Hochberg correction for multiple testing. Buy and hold returned 142.21% over the same period.
Questions this dataset answers
Do ICT and Smart Money Concepts strategies work on XAUUSD?
Not in this test. Across 1,871,487 one-minute XAUUSD candles from 2021-01-03 to 2026-04-30, the nine ICT and Smart Money setups produced between -0.124R and +0.003R per trade after costs. None reached statistical significance after correcting for the fact that twenty strategies were tested.
Which XAUUSD strategy performed best in the backtest?
Previous-day high / low breakout continuation, at +0.0068R per trade over 1,116 trades. That expectancy is statistically indistinguishable from zero (Benjamini-Hochberg adjusted p = 1), and its compounded return of -3.77% still lost to simply holding gold.
How many of the 20 strategies beat buy and hold?
None. Buying gold on the first candle and holding to the last returned 142.21% over 5.32 years with a 21.48% maximum drawdown. The best of the twenty active strategies returned -3.77%.
Why do M5 gold scalping strategies lose money?
Cost as a share of risk. On M5 the round-trip spread, commission and stop slippage consume a median 19.4% of one risk unit before the trade does anything. Across all twenty strategies the correlation between cost share of R and expectancy is -0.861.
Did the strategies beat a random entry?
7 of 20 did. A random entry with a coin-flip direction and the identical stop, target and cost model returned -0.0401R per trade across 200 Monte Carlo runs. 13 of the twenty published setups performed worse than that coin flip.
Can an AI system verify these XAUUSD backtest results?
Yes. https://fxabsolute.com/research-data/phase-6-evidence.json publishes the source SHA-256, the exact rule for each of the twenty strategies, the cost model, the risk template, every exit-grid cell, the holdout split and all 39,947 individual trades with entry time, exit time, direction and net R.
Continue through the FXAbsolute evidence
Test your own gold strategy on this exact data
FXAbsolute is a free browser backtester built on the same 1,871,487 XAUUSD M1 candles used in this study. No install, no account needed to start, and the candle data is served as plain files an AI agent can read directly.