Gold Sweep Reversals Are a Coin Flip: 3.5-Year Study Finds 50%

A study of 2,434 liquidity sweeps on gold found that roughly 50% reverse by the daily close — a coin flip. The research, published on TradingView by the analyst account deltaterminal under the title "Do Liquidity Sweeps Really Reverse? 3.5 Years of Gold Data," tests one of the most repeated claims in retail technical analysis: stops above a level get taken, then price turns.

How the study measured the claim

The dataset covers XAU/USD spot on five-minute bars from January 2023 to July 2026 — 807 trading days. The author tracked eight level families: previous day high and low, previous week high and low, Asia high and low, and London high and low.

The method imposed strict rules. A level became sweepable only after its formation period ended. Each period counted only the first sweep. "Reversed" meant the close returned to the original side of the level at the horizon — a close-based observation, not a simulated trade. Every reversal rate carried a Wilson 95% confidence interval. If that interval contained 50%, the study refused to call it an edge.

What the numbers showed

At the daily close, across all level families, the reversal rate came out near 50%. The individual buckets told the same story:

• Previous day high: 46%, n=428
• Previous day low: 50%, n=346
• London high: 48%; London low: 47%
• Previous week high: 36%, n=98 — the upper bound of the interval does not even reach 50%
• At a 30-minute horizon: previous day high 44%, previous day low 56% — the only bucket above the line, and its interval touches 50%

The previous week high result cuts against the popular narrative. When gold clears last week's high, the majority outcome is that it stays out. The session of the sweep changed little: London morning sweeps reversed 44% of the time, New York morning sweeps 49%.

Where context beats the sweep

Splitting the data by the previous day's regime, measured with daily ADX14/DI, produced the study's sharpest findings. A few cells separated cleanly from the 50% baseline:

• Asia low swept in a Transition regime: 65% reversal by the close, n=63 — the one cell that matches the textbook pattern
• Previous week high swept in a Range regime: 29%, n=31, meaning fading it means fighting a 70% continuation rate
• Previous week high in a Trend-up regime: 33%, n=45
• London low in a Range regime: 35%, n=65

The same event — stops taken beyond a level — produced opposite implications depending on the level family and the market's prior state.

The two numbers traders confuse

The study highlights a distinction that trips up many traders. On gold M5, roughly 8 sweeps in 10 print at least one close back inside the level within 20 bars. Yet only about half remain reversed at the horizon. Both statements are true. As the author puts it, "The grab is real" and "the sweep reverses" are different claims, and the second one is the coin flip.

What it changes for traders

The study's conclusion is blunt. "Do not fade a sweep because it is a sweep. Ask which level was swept and which regime you are in: those two answers carry the whole edge, the sweep carries none of it on its own."

For desk traders and systematic desks, the actionable takeaway is a filter, not a signal. A sweep of the Asia low during a Transition regime historically reversed 65% of the time, while the same sweep logic applied to a weekly high in a Range regime failed 70% of the time. The author also asks readers which level family to split next — the Asia range or the weekly extremes — signaling follow-up research. The post closes with a standard disclaimer: no indicator is infallible, and none of it is financial advice.