Short answer: no version passed, and the result has a shape worth more than the verdict. We tested five retracement depths on the same swings, the same waits, the same exits. The deeper the retracement, the worse the result, in a straight line, on both sides of the market. The two shallow depths we included as controls (0.382 and 0.5) beat all three OTE depths (0.618, 0.705, 0.786) on every chart from 1-minute to 4-hour. Shorting the deepest pullback, 78.6%, lost reliably enough to clear our statistical bar on three charts.
Read this first: what this test can and cannot tell you
- Gold only (XAUUSD), Deriv's feed, UTC candles, January 2017 to September 2026 (9 years 8 months).
- One precise, written definition, below, including how the swing is found, because "the leg" is the part every trader draws differently and it is where hindsight creeps in.
- Every trade pays the spread, at least 15 points, once per round trip.
- Entries are market orders at the next candle's open after price touches the level, not limit orders resting at it. This matters more here than anywhere else in the study; see "the test we threw away" below.
- The controls are built in. 0.382 and 0.5 are the same leg, the same wait, the same exit, at a depth nobody calls optimal. If the OTE zone is special, it has to beat them.
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 | Once per round trip, at least 15 points. Across this sample it averages 0.89 bp per trade, falling from 1.19 bp in 2017 to 0.37 so far in 2026 as gold's price rose. |
| Retracement depth (r) | How far back into the leg price pulls before you enter. 0.5 is halfway; 0.786 is a deep pullback close to the start of the move. |
| t | Distance from zero in units of the result's own noise. Under 2 is ordinary randomness. |
| The bar (3.71) | 240 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. |
| "Passed" | Cleared the bar and kept its sign across years and halves. It does not mean profitable, a reliable loss passes too, and 5 did. |
What we tested
The idea, in plain words. After a strong move, price usually pulls back before continuing. The claim is that the best place to join is a deep pullback, the 0.618 to 0.786 band, the "optimal trade entry", because you get a better price with a tighter stop and the move is still intact.
The rules, exactly.
- The leg (bullish): a confirmed swing low L, then a confirmed swing high H, with L the lowest low between them. A swing is confirmed only when 5 candles have printed on each side of it, so the leg is not known until 5 candles after the high, and the test does not let you trade a level you could not have drawn in real time.
- The level: P = H − r × (H − L), with r ∈ {0.382, 0.5, 0.618, 0.705, 0.786}.
- The wait: up to 50 candles from confirmation. Cancelled if price trades above H first. Skipped entirely if price already reached P before the swing could be confirmed.
- The trade: BUY at the next candle's open after the touch. Exit at the close 4, 12 or 24 candles later. Bearish mirrored.
- Charts: M1, M5, M15, M30, H1, H4, D1, W1.
- Grid: 8 charts × 5 depths × 3 holds × 2 directions = 240 versions.
The headline: deeper is worse, in order
Averaged across every version with at least 250 trades (about 60,000 to 175,000 trades per depth):
| Depth | Long: net per trade | Long: after gold's drift | Short: net per trade | Short: after gold's drift |
|---|---|---|---|---|
| 0.382 (control) | +0.13 bp | −0.05 | −1.58 bp | +0.39 |
| 0.500 (control) | +0.10 bp | −0.10 | −1.77 bp | +0.25 |
| 0.618 (OTE) | −0.24 bp | −0.45 | −1.44 bp | +0.56 |
| 0.705 (OTE) | −0.82 bp | −1.00 | −2.29 bp | −0.27 |
| 0.786 (OTE) | −0.87 bp | −1.04 | −3.95 bp | −1.91 |
Five depths, in order, on both sides. The two depths included purely as a comparison are the two best, and the depth sold hardest as "optimal" is the worst. There is no chart in the study on which the OTE band beat the control depths.
That is not proof the concept is backwards, the differences between neighbouring depths are individually small. It is, however, the opposite of what the idea predicts, measured consistently across hundreds of thousands of trades.
Results by depth on the charts people trade it on (exit after 12 candles)
| Chart | Depth | Direction | Trades | Win rate | Net (bp) | t | Passed? |
|---|---|---|---|---|---|---|---|
| M15 | 0.500 | long | 1,722 | 51.2% | −0.34 | −0.41 | no |
| M15 | 0.618 | long | 2,277 | 50.6% | +0.41 | +0.54 | no |
| M15 | 0.705 | long | 2,604 | 49.4% | −0.13 | −0.18 | no |
| M15 | 0.786 | long | 2,788 | 49.9% | −0.82 | −1.06 | no |
| M15 | 0.786 | short | 2,874 | 45.7% | −4.20 | −6.12 | yes, a reliable loss |
| M30 | 0.786 | short | 1,437 | 48.5% | −5.61 | −4.30 | yes, a reliable loss |
| H1 | 0.618 | long | 576 | 53.5% | −0.09 | −0.03 | no |
| H1 | 0.705 | long | 654 | 53.5% | −1.82 | −0.61 | no |
| H1 | 0.786 | long | 683 | 51.2% | −1.43 | −0.49 | no |
| H1 | 0.786 | short | 721 | 47.3% | −10.35 | −4.00 | yes, a reliable loss |
| H4 | 0.786 | short | 223 | 39.9% | −27.61 | −3.05 | too few trades |
5 of 240 versions passed the bar. All five were losses, and all five were the 0.786 short, M15, M30 (at all three holding times) and H1. Shorting the deepest pullback in a rising gold market is the one thing in this concept that did something consistently, and what it did was lose.
Note the win rates in that table: several versions win more than half their trades and still lose money. The losses are simply bigger than the wins.
Year by year: the 78.6% short, exit after 12 candles
15-minute chart:
| Year | Trades | Win rate | Net (bp) |
|---|---|---|---|
| 2017 | 288 | 45.5% | −2.82 |
| 2018 | 292 | 40.4% | −3.48 |
| 2019 | 303 | 43.6% | −3.25 |
| 2020 | 328 | 48.2% | −5.45 |
| 2021 | 271 | 45.0% | −3.88 |
| 2022 | 272 | 52.9% | −2.28 |
| 2023 | 285 | 48.1% | −1.94 |
| 2024 | 311 | 42.8% | −3.22 |
| 2025 | 309 | 42.7% | −8.43 |
| 2026 | 215 | 49.3% | −7.61 |
Hourly chart: −14.63, −1.55, −7.63, −20.72, +5.81, −7.89, −9.58, −10.13, −22.28, −14.41 bp (2017→2026), nine losing years out of ten, and the worst of them are the most recent.
The textbook plan: stop at the swing, target the extreme
The classic OTE trade risks the whole leg and targets the previous extreme: stop at 100% (the swing), target at 0% (the high you measured from), up to 200 candles. At a 0.705 entry that is roughly a 1:2.4 trade, and the naive prediction for a market with no pattern is that you reach the target about 1 − r of the time, 29.5%.
Computed from the actual fills rather than the idealised geometry (entries are at the next open, and the spread shifts both distances), the chance line for these trades is 34.2%. Gold delivered 31.0%.
| Depth | Chart | Direction | Trades | Target first (gold) | Chance gives | Shuffled gold | Gold − shuffled |
|---|---|---|---|---|---|---|---|
| 0.705 | M15 | long | 2,319 | 31.0% | 34.2% | 32.0% | −0.174 R |
| 0.705 | M15 | short | 2,305 | 27.4% | 30.6% | 27.9% | −0.070 R |
| 0.705 | H1 | long | 592 | 33.7% | 33.5% | 35.3% | −0.214 R |
| 0.786 | M15 | long | 2,460 | 24.3% | 27.7% | 24.4% | −0.145 R |
| 0.786 | M15 | short | 2,539 | 20.6% | 23.9% | 21.0% | −0.143 R |
| 0.500 | M15 | long | 1,585 | 50.3% | 52.1% | 51.8% | −0.061 R |
"Shuffled gold" is gold's own price jumps with the order scrambled, same volatility, same spread, no pattern at all.
Across all 48 comparable cells, gold came out −0.045 R behind the patternless baseline, and ahead in only 12 of them. The plan does not merely fail to beat chance; on this data it does slightly worse than the same plan run on a version of gold with the pattern taken out.
The test we threw away
Our first OTE test placed a limit order at the retracement level, the way it is actually taught. It produced a small consistent loss on gold, so we ran it on a patternless series to be sure the machinery was sound.
The patternless series produced a loss too, and a statistically strong one. That killed the test. A limit order simulated from candles carries a built-in bias: it only fills when price comes to it, so it systematically selects the moments just before price continues away from you, worth 0.2 to 0.8 bp per trade on a series with nothing in it at all. Any backtest that fills limits from candle highs and lows inherits that, and most do.
So the limit version was voided and rebuilt with market entries at the next open, which is the version on this page. We mention it because the lesson generalises: if your backtest fills limit orders from candles, its results carry a bias in the same size range as most of the edges people claim to find.
What this page does not say
- It does not say retracements do not happen. They do, the scanner found tens of thousands of them.
- It does not say fibonacci levels are meaningless as a map. It says that entering at the 0.618 to 0.786 band, with these rules, on gold, did not beat entering at 0.382 or 0.5, which is the only claim the "optimal" in "optimal trade entry" makes.
- It does not test OTE with a trend filter, a session filter, a higher-timeframe bias or a confirmation candle. Every one of those is a separate concept and needs its own pre-registered test.
How we tested
- The rules were written and dated before the scanner existed. Nothing was tuned afterwards.
- Swings are confirmed 5 candles late, and levels reached before confirmation are discarded, no level is traded that could not have been drawn at the time.
- 240 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: 0 survivors, but its strongest cell reached |t| 3.64 against a bar of 3.71. That is clean by our rule and we report it as close, because a control that nearly fires is information and hiding it would be dishonest.
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.