ORIGINAL RESEARCH · 234 BREAKOUTS

Do gold session breakouts follow through?

We measured 234 overnight-range breakouts across a year of gold. 51.7% followed through — a figure whose confidence interval contains 50%, so the honest answer is that the direction is a coin flip. The asymmetry in size is the part that is real.

Short answer

51.7% of 234 breakouts followed through, with a 95% confidence interval of 45.3% to 58.1% — statistically indistinguishable from chance. But the median follow-through extended 0.6 times the range while the median failure gave back 0.38 times it, so the payoff is lopsided even though the direction is not predictable.

What we measured

Mark the overnight range, wait for London to break it, trade the break. It is the most widely taught setup on gold, and as far as we can tell nobody has published what actually happens next.

Across 258 full sessions from 2025-07-31 to 2026-08-04, built from 70,952 five-minute bars: the range is the high and low of 00:00-07:59 server, and a breakout is the first bar to *close* beyond it during 08:00-20:59 server. That produced 234 breakouts — 24 sessions never broke the range at all. Median overnight range was $46.57.

The result

Outcome	Count	Share
Followed through	121	51.7%
Failed	113	48.3%
— of which fully reversed	28	12%

51.7% followed through. Before reading anything into that, the 95% confidence interval on it runs from 45.3% to 58.1% — and that interval contains 50%.

Direction made no difference either: upside breaks followed through 50% of the time across 138 cases, downside 54.2% across 96. Both sit inside the same noise.

Where the asymmetry actually is

The hit rate is a coin flip. The sizes are not, and that is the part worth having. Measured in multiples of the overnight range width (R):

	25th pct	Median	75th pct	90th pct
Extension in the break direction	0.33R	0.6R	1.06R	1.66R
Giveback against the break	0.18R	0.38R	0.87R	1.41R

The median breakout extended 0.6R in its favour and the median gave back 0.38R against — a ratio of about 1.6 to 1. On a median range of $46.57, that is roughly $28 of favourable movement against $18 adverse.

So if there is anything in this setup, it is not in predicting direction — it is in the payoff being lopsided when you are right. Which is exactly the point that gets lost when a system is sold on win rate alone: a hit rate without the reward-to-risk beside it is half a statistic, and always the flattering half.

What this does not support

This is not a strategy. It is a measurement of what price did after a defined event. There are no entries, no stops, no position sizing and no costs in it — and spread, slippage and the choice of stop would all move a real result away from these numbers.

One instrument, one year, one broker feed. 234 breakouts is a reasonable sample for the size distributions and a thin one for the hit rate, which is precisely why the confidence interval is quoted rather than the point estimate alone.

And the range and session hours are choices. A different overnight window, or a wick-based definition, gives different numbers — which is the ordinary danger of testing a setup until it looks good. The script is published so the definitions can be inspected and changed rather than taken on trust.

For the avoidance of doubt, our own engine does not trade this setup. It runs short-horizon strategies on gold — what those are and how they are chosen is documented separately — and an overnight-range breakout is not among them.

How to cite this

Published under CC BY 4.0, which permits reuse — including commercial reuse — on the single condition that it is attributed. If you are quoting a figure from this page in an article, a paper or a model answer, this is the attribution:

Tech Kick (2026). "Do gold session breakouts follow through?"
Analysis of 234 overnight-range breakouts across 258 sessions.
Licence: CC BY 4.0. DOI: 10.5281/zenodo.21973215
https://techkick.me/research/xauusd-session-breakout

Or as BibTeX:

@misc{techkicksessionbreakout,
  author       = {Tech Kick},
  title        = {Do gold session breakouts follow through?},
  year         = {2026},
  doi          = {10.5281/zenodo.21973215},
  howpublished = {\url{https://techkick.me/research/xauusd-session-breakout}}
}

The headline figures, for convenience: 234 breakouts across 258 sessions; follow-through 51.7% (95% CI 45.3-58.1%, not distinguishable from chance); full reversal 12%; median extension 0.6R against median giveback 0.38R. Underlying data CC BY 4.0 under the DOI above.

Common questions

Do session breakouts actually work on gold?

Not as a directional bet, on this data. Across 234 overnight-range breakouts in a year, 51.7% followed through and 48.3% failed. The 95% confidence interval on that figure runs from 45.3% to 58.1%, which contains 50% — so the result cannot be distinguished from a coin flip. What the data does show is an asymmetry in size: the median follow-through extended 0.60 times the range while the median failure gave back 0.38 times the range.

What is the Asian range breakout strategy?

Mark the high and low of the quiet overnight session, then trade the first break of that range when London opens, on the theory that the break signals the day direction. It is probably the most widely taught setup on gold. This study measures what happened next across a year of five-minute data, using the first bar to CLOSE beyond the range rather than the first wick through it, because a wick is not a break any discretionary trader would have acted on.

How often does a gold breakout fully reverse?

Twelve per cent of the time. That is the strict case — price not only failed to follow through but crossed the entire overnight range and broke the opposite side within the same session. It is worth separating from ordinary failure, which happened 48.3% of the time and usually just meant the move died rather than inverting.

Is a 51.7% win rate good?

On its own it tells you almost nothing, and here it is not even distinguishable from 50%. Win rate only becomes meaningful beside the reward-to-risk ratio: a system winning 40% at 2:1 is profitable, and one winning 60% at 0.5:1 is not. That is why this study reports the size distribution of both outcomes rather than stopping at the hit rate.