Volume profile POC pullback after a break of structure: does it work?
A viral short-form video teaches this: in a downtrend, mark the last break of structure, draw a fixed-range volume profile from the lower high to the lower low, find the point of control (POC, the price with the most volume), wait for price to retrace into it with some breathing room, then short and aim for the lower low. Longs are the mirror image.
The exact rules we tested
- Swings: pivots with 3 or 5 bars on each side. A down setup needs a lower high, a lower low, and a candle close below the prior low (the break of structure; wicks do not count).
- Volume profile: 40 price bins across the leg from the lower high to the lower low, each bar's volume spread across its range. POC is the centre of the busiest bin.
- Entry: a limit order at the POC, or at the POC minus 10% of the leg for the "breathing room" version. Stop at the lower high, target at the lower low, order cancelled after 48 bars.
- Data: 8 Binance USD-M perpetuals (BTC, ETH, SOL, BNB, XRP, DOGE, ADA, LINK) on 15-minute and 1-hour bars. Costs 0.04% taker fee and 0.02% slippage per side; funding payments are not in the headline numbers (their size is reported below). Results are in R, where 1R is the distance from entry to stop.
- Splits: development 2021-01-01 to 2023-12-31, test 2024-01-01 to 2026-09-30. All 8 configurations (3 or 5 bar swings, 15m or 1h, two entries) were declared in a pre-registration file committed together with the results, so the order of writing and running cannot be independently shown.
Results
| Configuration (test split) | Trades | Win rate (95% CI) | Expectancy, net (test) | Profit factor | Expectancy, net (dev) |
|---|---|---|---|---|---|
| 3-bar swings, 15m, POC | 8,107 | 49.8% (48.7 to 50.9) | -0.141R | 0.76 | -0.177R |
| 3-bar swings, 15m, POC with room | 7,077 | 51.8% (50.6 to 52.9) | -0.139R | 0.75 | -0.153R |
| 3-bar swings, 1h, POC | 2,265 | 43.6% (41.5 to 45.6) | -0.169R | 0.73 | -0.061R |
| 3-bar swings, 1h, POC with room | 1,978 | 48.2% (46.0 to 50.5) | -0.122R | 0.78 | -0.034R |
| 5-bar swings, 15m, POC | 5,511 | 48.7% (47.4 to 50.0) | -0.138R | 0.76 | -0.146R |
| 5-bar swings, 15m, POC with room | 4,875 | 51.4% (50.0 to 52.9) | -0.132R | 0.76 | -0.124R |
| 5-bar swings, 1h, POC | 1,524 | 45.0% (42.5 to 47.6) | -0.131R | 0.78 | -0.083R |
| 5-bar swings, 1h, POC with room | 1,254 | 48.6% (45.8 to 51.4) | -0.087R | 0.84 | -0.039R |
All eight versions lost money after modelled costs in both the development and the test split. The smallest p-value for "expectancy above zero" in the test split was 0.996, and the smallest Benjamini-Hochberg q across all controls was 1.00. Quote as: the tested configurations did not pass on these data and assumptions (after modeled taker fees and slippage; funding excluded).
What happened
None of the eight tested versions passed. Before costs, expectancy in the test split ran from -0.070R to +0.011R a trade, so there is no gross edge for fees and slippage to eat. The version that looked best on the development data (3-bar swings, 1h, POC with room, -0.034R) carried to the test split at -0.122R, so picking the best configuration did not survive out of sample. Across all eight configurations and six years, 47 of 48 config-years were negative, and longs and shorts were equally bad on the test split (-0.137R and -0.137R). Funding payments, tested separately, moved expectancy by no more than 0.0012R a trade.
Why a win rate near 50% does not rescue it
A win rate around 50% sounds fine until you read the profit factor. With the stop at the lower high and the target at the lower low, the average win is not large enough to cover the average loss plus fees and slippage, so the profit factor sits between 0.73 and 0.84 on the test split. Test win rates ran from 43.6% to 51.8%, and several versions win about half their trades and still lose 0.087R to 0.169R each time.
Does the POC add anything?
Not in these tests. A control that uses the same swings but enters at the 50% retracement lost less than the POC entry in 8 of 8 test configurations, so the volume profile did not improve the entry. We also withdraw an earlier claim about direction. A first run suggested that the strategy's direction beat a coin-flip direction in two configurations (p-values of 0.002 and 0.002). That came from a bug: the coin-flip control could be charged a stop that was hit before its entry. With the bug fixed the same two comparisons give p = 0.33 and 0.48, and across all eight configurations the range is 0.22 to 0.96. There is no evidence that the direction carries information, and no claim that it beats a coin flip. A control with the same stop and target distances entered at random bars gave a minimum q of 1.00 for beating the strategy in the wrong direction; none of the controls separated from the strategy.
Controls
Three controls were run per configuration: entering at the 50% retracement, a coin-flip direction at the same fill bars, and a random-bar entry with the same stop and target distances. p-values were corrected with Benjamini-Hochberg across the tests in each split. The pre-registered verdict needed a configuration with positive test expectancy at q below 0.05 that also beat all three controls. no tested configuration passed. Intervals use week-block resampling, because trades on different coins in the same week are not independent.
Limits
A second run (v2) was made after an independent audit of the first and fixed several execution and labelling defects. It changed some counts slightly and changed no verdict. This tests a mechanical version. A discretionary trader who only takes the visually clean setups cannot be tested this way. The clip's "breathing room" is vague, so we tested the POC and the POC minus 10% of the leg. Fill assumptions are simplified: fees are charged as taker fees even on limit fills and we did not test more favourable assumptions, so the results are after modelled fees and slippage only. Positions across coins are correlated, so the effective sample is smaller than the trade count. As a supplement we ran gold (Dukascopy XAUUSD, 15 minute) using tick volume, which is a proxy rather than exchange volume, with no fee. The first gold download was defective; after a clean refetch all eight gold configurations had negative test expectancy (-0.144R to -0.050R on 107 to 297 trades each, best q = 0.92). The earlier mixed signs on gold came from the defective data.
General information only. It doesn't consider your objectives, finances or needs. Tickfloor holds no financial services licence and never places trades.
Does the volume profile POC strategy work?
These versions did not pass our tests. All 8 pre-registered versions lost money after modelled costs in both the development split (2021-01-01 to 2023-12-31) and the test split (2024-01-01 to 2026-09-30) on 8 Binance USD-M perpetuals. Test win rates were 43.6% to 51.8% and expectancy was -0.087R to -0.169R a trade.
Is the point of control better than a plain 50% retracement?
In these tests, no. Entering at the 50% retracement lost less than the POC entry in 8 of 8 test configurations, so the volume profile did not improve the entry.
Does the direction of the trade beat a coin flip?
We found no evidence of it. An earlier suggestion came from a control bug; after the fix the comparison p-values range from 0.22 to 0.96.
Did it work on gold?
No. After a clean data refetch all eight gold configurations had negative test expectancy (-0.144R to -0.050R), none significant. Gold was a supplement, using tick volume, and was not part of the verdict.
Is this financial advice?
No. This measures a publicly claimed trading rule, not a recommendation. General information only, not personal advice.
How we set the rules, costs and pass bar before running anything: the Tickfloor testing standard. The wider record is on the evidence page.