The Asian session is usually gold’s quietest. Its high and low become obvious reference points, and stop orders collect just beyond them: buy stops above the Asian high (buy-side liquidity) and sell stops below the Asian low (sell-side liquidity). The common claim is that London “sweeps” one side before the real move. Here is how often that actually happened, with the sample behind each figure.
The results
- Almost every day: one or both sides were taken on 113 of 116 days. Only 3 days stayed inside the Asian range all day.
- Both sides: on 29 days (29 of 116 days · 95% range 18–34%) price took the high and the low, so a first sweep is no promise of direction.
- London first: of the 113 days with a break, London took the first side on 103 and New York on 10.
- Back inside fast: 109 of 142 breaks closed back inside within an hour. Breaks are counted per side, so one day can add two; breaks on the same day are related, which makes the true range a little wider than shown.
- Typical overshoot: the median distance beyond the level before the close back was about $3 (30 pips).
What a sweep looks like
The detail that matters is the close. A wick beyond the level that closes back inside is a sweep. A close beyond the level that holds is a break, and in this sample that happened in roughly one case in four (33 of 142 breaks).
From a sweep to a confirmed idea
In our method a level being taken is where we start watching, never the entry. The sequence is fixed: 1 · sweep of the level; 2 · CHoCH, a candle body closing beyond the external swing in the new direction (a wick does not count); 3 · retest of the broken level or the order block of the CHoCH leg; only then is the idea confirmed, and the order block of the CHoCH leg is the zone to study. No break and retest, no trade. This is a method for study, not a call to buy or sell.
This is why a 77% “close back inside” rate is not a trading rule on its own. It says nothing about how far price travels next, and it ignores spreads and costs. The confirmation steps filter out most sweeps that go nowhere. The full method is in our lessons on CHoCH vs BOS and order blocks.
When the day’s high and low print
We also logged the hour of each day’s high and low (232 extremes on 116 days). About 27% printed in the first three hours after the 18:00 New York open, and about 25% between 09:00 and 12:00 New York, around the US data releases and the cash open. Quiet hours rarely set the day’s extremes. For session times in your own time zone, see our sessions and killzones guide.
What this does and does not mean
- It describes how gold behaved in one five-month stretch in which gold trended down. A different market could give different numbers.
- There is no baseline here for “any level at a random time”, so these rates show how often things happened, not an edge over chance.
- High-impact news (CPI, NFP, FOMC) can push straight through a level. Our scanner stands aside from 60 minutes before to 45 minutes after marked news.
We will re-run the study as more clean data arrives and update this page with the new date. The live version of these levels, built with the same rules, is on the Gold Liquidity Map.
How we measured it
- Data: 15-minute XAU/USD bars from our scanner’s price feed, 116 trading days from 30 April to 9 October 2026, every day with clean history at the time of writing. No days were picked or dropped for their result.
- Clock: New York time; the gold trading day starts at 18:00. New York was on daylight time (UTC−4) in this period, so 18:00 New York was 03:00 Pakistan time.
- Asian range: the high and low between 20:00 and 24:00 New York (00:00–04:00 UTC, 05:00–09:00 Pakistan time in this period).
- Taken: price trades beyond the level later the same trading day. Closed back: a 15-minute candle closes back on the original side within four candles (one hour).
- Ranges: 95% Wilson score intervals. They assume independent events; levels on the same day are related, so the true uncertainty is somewhat wider.
- Same rules as our tool: the Gold Liquidity Map uses identical code, so these numbers describe exactly what it shows.