Safu Crypto Free researchLiquidity sweeps: does price really reverse after one?
Almost every trader has heard the story: price grabs the stops above a high, snaps back, and a reversal follows. A research team put that story to the test on 37,162 sweeps across 7 futures markets, and compared every real level with a fake one built to look the same. Here is what they found, in plain English, and how you can use the same testing mindset on your own trading.
After a sweep, price reversed 49.9% of the time. After a sweep of a matched fake level, 50.0%. That is a coin flip, and none of the 97 market and level combinations beat the fakes.
confirmation combinations made money on real levels at a 1 to 1 target. On fake levels, up to 28 did. Testing hundreds of filters will always produce a few lucky winners.
The one real effect runs the other way. At the first touch, real previous day highs and lows were run through 56.0% of the time. Fake levels: 50.1%.
Trading volume in the first second at a real level, against 5.5× at a fake one. The stops really are there, and they are gone in a second.
What a liquidity sweep is
A liquidity sweep happens when price pushes through an obvious high or low and then closes back inside the range. The popular story: stop orders sit just beyond those levels, price runs them, the stops get filled, nobody is left to push further, and price turns around.
The study tested the levels most traders mark: the previous day high and low, the Asia session and London session highs and lows, and the prior week high and low.
| Level tested | Sweeps found | First touches |
|---|---|---|
| Previous day high / low | 6,009 | 13,587 |
| Asia session high / low | 8,546 | 19,087 |
| London session high / low | 8,120 | 17,768 |
| Prior week high / low | 1,043 | 2,460 |
| All four | 23,718 | 52,902 |
Markets: NQ, ES, YM and RTY (stock indexes), GC (gold), SI (silver) and CL (crude oil). Sweeps were found on 5 minute candles.
What each sweep is compared with
To know whether a level matters, you need something to compare it with. So for every real level the team built a fake one: a price that is not a key level, but matches it in every other way.
Both were run through the exact same code. If real levels matter, they have to beat the fakes.
The bar for proof
Before looking at any result, the team wrote down what would count. A finding had to:
- Hold on the data the study was built on (2019 to 2024), and again on data it had never seen (2025 and 2026)
- Survive a correction for running many tests, because test enough things and some will look good by luck
- Beat what the same test finds on randomly shuffled prices
The authors also disclose that part of the 2025 and 2026 data had been seen by earlier studies, so that check is a second look, not a first one.
Does price reverse after a sweep?
From the close of the sweep candle, the team set a target one ATR away in the reversal direction and a stop one ATR back through the level. Then they asked: which gets hit first? Fifty percent is a coin flip.
ATR is the average range of a candle. It works as a ruler, so results are comparable in quiet and wild markets.
| Level group | Real | Fake | Gap |
|---|---|---|---|
| Previous day high / low | 50.4% | 49.6% | 0.8 higher |
| Asia session high / low | 50.4% | 50.5% | 0.1 lower |
| London session high / low | 49.3% | 50.1% | 0.8 lower |
| Prior week high / low | 49.0% | 48.8% | 0.2 higher |
| All four | 49.9% | 50.0% | 0.1 lower |
The gap across all sweeps is 0.1 points, and the data was large enough to spot any edge above 1 point. This is a measured zero, not a shortage of data. Real sweeps did not reverse more often than fake ones, in any market the team checked.
Does anything make a sweep reverse?
Most traders say a sweep needs confirmation before you take it. So the team tested the ones that get taught. Each was traded the same way: enter on the next 1 minute candle, stop just past the wick, target the same distance as the stop. Then the exact same trades were run on fake levels.
| Confirmation | Wins on real | Per trade, real | Per trade, fake |
|---|---|---|---|
| No confirmation at all | 45.4% | loses 0.23R | loses 0.25R |
| Break of structure | 45.2% | loses 0.14R | loses 0.11R |
| Fair value gap | 45.5% | loses 0.14R | loses 0.07R |
| Volume surge | 44.1% | loses 0.25R | loses 0.25R |
| Two levels stacked | 46.9% | loses 0.20R | loses 0.20R |
R is the amount you risk on one trade. Losing 0.23R means losing about 23% of your risk per trade on average, after costs. A selection of the tested filters is shown here.
- Every confirmation lost money on real levels. The best ones, break of structure and fair value gap, lost less, but they did even better on fake levels.
- Stacking filters did not rescue it. Of 263 combinations, 1 made money on real levels. On fake levels, up to 28 did.
- That is what over testing looks like. Try hundreds of filters on random prices and a few will always look like winners.
Imagine testing any trading idea before you risk a single dollar.
That is exactly how the Safu AI Trading Bootcamp works. You bring the idea. AI builds the test. And the ideas that survive can run as your own bot, 24/7, while you live your life.
The levels are real. Price runs through them.
If sweeps do not reverse, do these levels mean anything at all? To find out, the team stopped waiting for a sweep and looked at the first time price reaches each level. From there, does it break through or bounce back? Both were measured at the same distance, one ATR, at real levels and at fake ones.
If levels were just lines on a chart, real and fake would look the same. They do not, but the difference points the opposite way to what gets taught.
| Broke one ATR through first | Real | Fake | Gap |
|---|---|---|---|
| Previous day high / low | 56.0% | 50.1% | 5.9 higher |
| Asia session high / low | 53.3% | 49.0% | 4.3 higher |
| London session high / low | 55.4% | 51.7% | 3.7 higher |
| Prior week high / low | 61.9% | 50.3% | 11.6 higher |
52,902 first touches across seven markets. The effect kept its direction in every market and every year the team checked.
- The stops are real. A volume burst that big is stop orders being hit all at once.
- They get run, not respected. About 70.5% of that burst traded in the break direction.
- It fades with age. The effect was strongest on the same day and close to zero after 20 days.
- Still not a free trade. Betting on the break with an equal stop and target lost money after costs. The edge is smaller than what each trade costs.
What gets taught, next to what the data shows
Sweeps do not reverse more than fake sweeps. The levels are real, and they get run through.
Price reversed after a sweep no more often than after a sweep of a matched fake level, in any market, at any level, with any confirmation the team tested. At the first touch these levels were broken more often than empty price. The stops really are sitting there, and fading the level by default means trading the wrong side of them.
What this means for a crypto trader
One honest caveat first. This study covers CME futures. Crypto trades around the clock and has its own behavior, so the exact numbers may not carry over. The lesson is not that sweeps are useless. The lesson is that a pattern you can see on a chart is not an edge until it survives a fair test.
That is how we treat sweeps in Safu AI too. A sweep never counts just because it looks like one. It has to pass strict rules in the right market condition, and every rule gets tested. Here is the loop to use on any idea:
- Write the idea as strict rules. If two traders could draw it differently, it is not a rule yet.
- Compare it with a control. Test it against random or fake setups, not just against zero.
- Check it on data it never saw. A rule that only works on the data you built it on is a story, not an edge.
- Count the real costs. Fees and slippage decide more strategies than most traders expect.
- Only then automate it. Let a bot run what has earned the right to run.
You do not need to write code to do any of this. In the bootcamp, AI handles the code and you learn the thinking: how to ask the right question, how to read the result honestly, and how to turn the winners into a bot.
Stop guessing. Start testing.
You are not late. Join today, get every recorded class, and build your own AI trading system step by step.