Short answer: no version passed. On the 1- and 5-minute charts CRT lost reliably enough to clear a demanding statistical bar, 12 versions did, every one a loss. The plan most people trade it with hits its target 62 to 71% of the time, which sounds excellent until you learn that its stop-and-target geometry gives 74 to 80% by chance alone, and that a version of gold with every pattern shuffled out of it hits the same targets just as often.
Read this first: what this test can and cannot tell you
- Gold only (XAUUSD), Deriv's feed, candles on the UTC clock, January 2017 to September 2026 (9 years 8 months).
- One precise, written definition, below. CRT is taught with several different candle counts; we tested the canonical three-candle version.
- Every trade pays the spread, at least 15 points, from the broker's own per-candle record, once per round trip.
- Entries are market orders at the next candle's open.
- This page reports a second test. Our first one was thrown out by our own control, the story is at the bottom, because it explains more about testing than the result does.
How to read the numbers (skip if you already know)
| Term | What it means here |
|---|---|
| bp (basis point) | 0.01% of price. At $4,300 gold, 1 bp ≈ $0.43 per ounce. |
| The spread you pay | Charged once per round trip, at least 15 points, from the broker's own record. Across this sample it averages 0.89 bp per trade, and it has been shrinking as gold has risen: 1.19 bp in 2017, 0.79 in 2023, 0.37 so far in 2026. |
| R | One unit of risk. A trade risking $100 that makes $200 returned +2R. Results for plans are quoted in R, not money, so position size drops out. |
| The barrier law | With a stop and a target, a market with no pattern at all hits the target first with probability risk ÷ (risk + reward). A tight stop and a near target win most of the time and still lose money. This number is computed per trade from the actual distances, not assumed. |
| t | Distance from zero in units of the result's own noise. Under 2 is ordinary randomness. |
| The bar (3.58) | 144 versions were tried, so the threshold rises to match (Bonferroni, two-sided 5%). |
| Alpha | What is left after subtracting gold's own move over the same holding time in the same year. |
What we tested
The idea, in plain words. Three candles tell a story: an anchor that sets a range, a raid that pokes out of it and closes back inside, and a confirmation that closes beyond the raid. The claim is that the raid was a stop-hunt and the confirmation is the real move starting.
The rules, exactly. A, R and C are consecutive candles.
- Bullish:
R.low < A.low(the raid takes the anchor's low),A.low < R.close < A.high(it closes back inside), andC.close > R.high(the confirmation closes beyond the raid). Bearish mirrored. - The trade: BUY at the open of the candle after C. Exit at the close 4, 12 or 24 candles later.
- Filters: none (F0); the raided extreme within 10 bp of the previous day's high or low (F1); within 15 bp of the previous week's (F2).
- Charts: M1, M5, M15, M30, H1, H4, D1, W1.
- Grid: 8 charts × 3 filters × 3 holding times × 2 directions = 144 versions.
Results by chart: no filter, exit after 12 candles
| Chart | Direction | Trades | Win rate | Net per trade (bp) | t | Passed the bar? |
|---|---|---|---|---|---|---|
| M1 | buy | 113,012 | 42.9% | −0.91 | −30.91 | yes, a reliable loss |
| M1 | sell | 114,115 | 42.1% | −0.98 | −33.62 | yes, a reliable loss |
| M5 | buy | 25,289 | 46.0% | −1.05 | −7.85 | yes, a reliable loss |
| M5 | sell | 25,147 | 45.2% | −1.13 | −8.27 | no (alpha fails) |
| M15 | buy | 8,605 | 48.3% | −0.33 | −0.85 | no |
| M15 | sell | 8,527 | 45.6% | −1.61 | −4.19 | no (alpha fails) |
| M30 | buy | 4,238 | 50.9% | +0.60 | +0.77 | no |
| M30 | sell | 4,242 | 46.8% | −1.92 | −2.49 | no |
| H1 | buy | 2,106 | 51.8% | +1.04 | +0.65 | no |
| H1 | sell | 2,106 | 45.9% | −2.08 | −1.30 | no |
| H4 | buy | 544 | 54.0% | +7.29 | +1.23 | no |
| H4 | sell | 518 | 45.6% | −7.17 | −1.10 | no |
| D1 | buy | 106 | 54.7% | +53.66 | +1.86 | too few trades |
| D1 | sell | 92 | 48.9% | −9.52 | −0.33 | too few trades |
| W1 | buy | 20 | 70.0% | +333.24 | +1.80 | too few trades |
| W1 | sell | 21 | 19.0% | −588.40 | −3.89 | too few trades |
12 of 144 versions passed the bar, and all 12 were losses, all on the 1- and 5-minute charts. On the random control series: 0 of 144, maximum |t| 2.72.
The fast charts are where CRT is most often traded and where it did worst: about −0.9 to −1.1 bp per trade, which is roughly the spread. The pattern itself is close to neutral; the cost is what you keep.
The filters did not help
| Chart | Direction | No filter | At yesterday's high/low | At last week's high/low |
|---|---|---|---|---|
| M15 | buy | −0.33 bp (8,605) | −1.47 bp (1,065) | −0.55 bp (398) |
| M15 | sell | −1.61 bp (8,527) | −0.67 bp (1,186) | +0.18 bp (549) |
| H1 | buy | +1.04 bp (2,106) | −2.01 bp (308) | −10.46 bp (117) |
| H1 | sell | −2.08 bp (2,106) | −0.65 bp (327) | −1.50 bp (163) |
Requiring the raid to happen at a "proper" level, the classic refinement, cut the sample by 85 to 95% and moved the numbers in no consistent direction. Nothing here clears anything.
Year by year: bearish CRT on the 15-minute chart, exit after 12 candles
| Year | Trades | Win rate | Net per trade (bp) |
|---|---|---|---|
| 2017 | 835 | 44.1% | −2.44 |
| 2018 | 888 | 45.6% | −1.73 |
| 2019 | 870 | 44.1% | −2.64 |
| 2020 | 895 | 44.4% | −2.48 |
| 2021 | 848 | 47.1% | −0.50 |
| 2022 | 908 | 47.4% | −1.63 |
| 2023 | 885 | 47.3% | −0.44 |
| 2024 | 886 | 43.2% | −3.08 |
| 2025 | 883 | 43.9% | −2.76 |
| 2026 | 629 | 49.8% | +2.82 |
Nine losing years, then a positive 2026 on a partial year. This is what a genuinely negative rule looks like, not a collapse, just a steady bleed roughly the size of the cost.
The plan everyone actually trades, and the trap inside its win rate
The textbook CRT plan: stop at the raid's extreme, target the far side of the anchor, ten candles maximum. Because the anchor is usually not much bigger than the raid, this produces a tight stop and a close target, and that makes the win rate high by construction.
Below: gold, versus the same plan run on shuffled gold, a series built from gold's own price jumps with the order scrambled, so it has gold's volatility, gold's spread, gold's gaps, and no pattern whatsoever.
| Chart | Direction | Trades | Target first (gold) | Chance gives | Shuffled gold | Result (gold) | Gold − shuffled |
|---|---|---|---|---|---|---|---|
| M1 | buy | 33,140 | 62.7% | 80.1% | 51.2% | −0.214 R | +0.070 R |
| M1 | sell | 68,272 | 45.8% | 74.2% | 38.0% | −0.392 R | +0.051 R |
| M5 | buy | 11,435 | 67.6% | 79.5% | 63.2% | −0.151 R | +0.005 R |
| M5 | sell | 16,263 | 59.6% | 75.3% | 59.3% | −0.211 R | −0.032 R |
| M15 | buy | 4,609 | 70.2% | 78.5% | 68.0% | −0.108 R | 0.000 R |
| M15 | sell | 5,744 | 65.5% | 75.1% | 66.9% | −0.126 R | −0.025 R |
| M30 | buy | 2,374 | 70.9% | 77.9% | 70.9% | −0.089 R | −0.018 R |
| M30 | sell | 2,831 | 65.5% | 74.6% | 68.9% | −0.130 R | −0.045 R |
| H1 | buy | 1,292 | 69.4% | 77.9% | 72.5% | −0.106 R | −0.049 R |
| H1 | sell | 1,334 | 64.6% | 73.4% | 68.9% | −0.137 R | −0.040 R |
Three things to take from that table:
- A 70% win rate here is a losing strategy. The target sits closer than the stop, so chance alone would give 77.9%. Winning seven times in ten while losing more on each loss than you make on each win is exactly what −0.089 R per trade looks like.
- Shuffled gold wins about as often. On the 30-minute chart it hits the same targets 70.9% of the time, identical, with no pattern in it at all. Whatever produces the high strike rate, it is the geometry, not the three candles.
- Across all 24 comparable cells, gold beats the patternless baseline by −0.009 R on average, and in only 10 of 24 cells at all. The concept does not beat a version of itself with the pattern removed.
Both gold and the baseline fall short of the chance line in the same way, by roughly the same amount. That shortfall is the mechanical cost of tight stops, spread, and candles that touch stop and target in the same minute (counted as the stop, because we cannot see the order inside a candle and assuming the good one is how backtests flatter themselves).
The test we threw away, and why we are telling you
Our first CRT test used the same plan but with a take-profit limit at the target. It produced a small, consistent loss on gold. It also produced a small, consistent loss on the random control series, about −0.05 R per trade, on 10 versions, past the significance bar.
When the control "finds" something, the control is telling you the machinery is broken, not the market. A take-profit limit that fills exactly at your target when price jumps clean through it silently forfeits the jump, and tight targets make that happen constantly. So the first CRT test was voided, struck off the record, not reported as a result, and rebuilt with clock exits and market entries. That rebuilt version is this page.
We kept the void on record and we mention it here because most published backtests never run a control that can embarrass them. Ours did, and it cost us a day.
What this page does not say
- It does not say the three-candle sequence is meaningless as a way of reading a chart.
- It does not test CRT combined with a session time, a higher-timeframe bias or a trend filter. Each is a separate concept needing its own pre-registered test, not a variation added after seeing these results.
- It does not test other instruments.
How we tested
- The rules were written and dated before the scanner existed. Nothing was tuned afterwards.
- Closed candles only; market entry at the next open; one position at a time per version.
- 144 versions, so the bar rises to match (Bonferroni). A version must also hold its sign in ≥70% of years, in both halves, and survive removal of gold's drift.
- The identical grid on a random price series: clean, 0 survivors, maximum |t| 2.72. The plan is judged instead against shuffled gold, which is a harder and fairer baseline for a stop-and-target rule.
Reproduce it: every rule is written out in full above, so anyone with gold price data can rebuild this test and check our numbers. The candles are the broker's and are not ours to redistribute.