Does the liquidity sweep strategy actually work?

Yes — but barely, and only with a wide target. Tested on 2.54 years of XAUUSD M15 data with 30 points round-trip spread subtracted from every trade, the liquidity sweep fade produced a profit factor of 1.12 from 173 trades at a 25.4% win rate. That is a genuine edge, but a thin one that dies entirely at lower reward-to-risk.

Based on 60 000 M15 XAUUSD bars — 2024-01-05 to 2026-07-21 (2.54 years of real market data), with 30 points round-trip spread subtracted from every trade.

What we actually tested

The setup is the standard smart-money liquidity sweep: price pushes beyond an obvious swing high or low — where retail stop orders cluster — then closes back inside the range. The fade takes the opposite side of that raid, entering on the rejection with a stop beyond the sweep wick. If you want the mechanics rather than the results, we cover them in liquidity and sweeps.

We ran it over 60 000 M15 bars of XAUUSD — 2024-01-05 to 2026-07-21, which is 2.54 years and the maximum M15 history our broker serves. Every trade was charged 30 points round-trip spread subtracted from every trade, because a strategy that only works before costs does not work.

Two filters were active throughout: entries had to agree with the prevailing trend, and only the high-volume session window was traded. Both were included because we had already measured them as improvements rather than assumed it.

The headline result

Across 173 trades the strategy won 25.4% of the time with a profit factor of 1.12, an expectancy of $0.37 per trade and a total of $63.53 on 0.01 lots. Maximum drawdown was $65.65.

A profit factor of 1.12 means that for every dollar lost, $1.12 was gained. It is positive, it survived costs, and it is nothing like the numbers sold in most strategy courses. That is what a real, modest edge looks like.

The finding that mattered most: reward-to-risk decides everything

The same entries were profitable or catastrophic depending purely on where the target sat. We swept reward-to-risk from 2.0 to 5.0 and re-ran the entire dataset each time, net of costs:

Reward:RiskTradesWin rateProfit factorExpectancyTotal
2:117432.2%0.87$-0.35$-61.18
2.5:117428.7%0.96$-0.11$-19.96
3:117326.6%1.02$0.07$12.3
3.5:117325.4%1.12$0.37$63.53
4:117323.1%1.12$0.36$62.58
5:117219.8%1.13$0.41$70.83
Identical entries, only the target moved. At 2:1 the strategy loses money; from 3:1 it turns positive. This is why reward-to-risk beats win rate.

Why lower targets destroyed it

At 2:1 the strategy won 32.2% — the highest win rate in the whole table — and still lost $61.18. At 5:1 it won only 19.8% and made the most money.

That is the arithmetic of expectancy doing its work. At 2:1 you need to be right 33.3% of the time just to break even before costs; the strategy managed 32.2%, which is under water once spread is charged. At 3.5:1 the break-even requirement falls to about 22%, and 25.4% clears it.

We settled on 3.5:1 rather than the nominally best 5:1 because 3.25 through 4.0 formed a stable plateau of similar results. Choosing the middle of a plateau is far more robust than choosing a lone peak, which is usually curve fitting in disguise.

What it costs you psychologically

A 25.4% win rate means roughly three of every four trades lose. In this dataset the worst run was 11 consecutive losses. That is not a tail risk we are warning you about — it is in the record.

Most traders cannot sit through eleven losses in a row without changing something: widening a stop, skipping the next signal, doubling size to recover. Any of those breaks the edge. The strategy's profit factor of 1.12 only exists if you take every signal, including the ones after a brutal streak. The discipline required is covered in trade review.

How thin the edge really is

Expectancy came in at $0.37 per trade on 0.01 lots. Round-trip spread on XAUUSD at that size is roughly $0.30. In other words, costs consume a large share of the gross edge — which is exactly why we model them, and why any backtest that omits spread should be treated as fiction.

It also means execution quality matters enormously. A wider-spread broker, slippage on entry, or trading during thin hours can flip this from marginally profitable to marginally unprofitable. The edge is real but it has almost no margin for sloppiness.

The honest caveats

173 trades is a modest sample. It is enough to reject an obviously broken idea but not enough to be confident about a thin one. A profit factor of 1.12 on 173 trades could plausibly be luck.

This is a historical simulation, not a live track record. Simulated fills are optimistic by nature: we assume the stop is hit first when a single bar spans both stop and target, but real slippage during fast moves can be worse than modelled.

And it is one instrument over one 2.5-year window that happened to contain a strong gold trend. Nothing here proves the behaviour persists.

So should you trade it?

The defensible conclusion is that the liquidity sweep is worth forward-testing, not worth betting on. It is the only one of three strategies we tested that survived costs at all — the other two lost money, which we document in does the opening range breakout work on gold.

If you do trade it: use a target of at least 3:1, expect to be wrong three times out of four, size so that eleven consecutive losses is survivable, and track your own expectancy rather than trusting ours. Our numbers are a starting hypothesis, not permission.

Want the method behind the numbers?
GoldSwift teaches gold structure, liquidity, risk and review — the same framework these tests come from. Start free. See membership plans →

Frequently asked questions

Is the liquidity sweep strategy profitable?

In our test on 2.54 years of XAUUSD M15 data it produced a profit factor of 1.12 net of spread — profitable, but a thin edge that requires a reward-to-risk of at least 3:1 to survive.

What win rate does the liquidity sweep have?

25.4% at a 3.5:1 target. Win rate rises to 32.2% at 2:1 — but the strategy loses money there, because the target is too small.

How many losses in a row should I expect?

Our dataset contained a worst streak of 11 consecutive losses. Size your risk so that is survivable.

Does it work without spread costs?

It looks considerably better without costs, which is exactly why excluding them is misleading. Every figure quoted here has spread subtracted.

Related, with the numbers