Asia Doesn't Predict London. We Checked 821 Sessions

It’s intuitive. It’s everywhere. And on Gold, across every session we could measure, it’s not true.

We checked 821 trading sessions of XAU/USD from our 2023–2026 pipeline and asked the simplest possible version of the question: when Asia closes in a given direction, how often does London continue in that direction?

The answer is 49.3%, with a 95% confidence interval of [45.9–52.7]. Dead center on the coin flip. Not “a small edge that needs size to exploit” — nothing. This article walks through the numbers, dismantles the one seductive statistic that seems to rescue the theory (the famous 74.8%), and makes the case for why a null result like this one is worth more to a trader than most “edges” being sold.

Method

• Data: XAU/USD spot (Dukascopy), n = 821 days, 2023–2026 pipeline window.
• Question: does the direction of one session (up or down over the session) predict the direction of the next?
• Baseline drift: Asia closed up on 53.2% of days in the sample — Gold spent much of this window in an uptrend, so raw up-rates carry that drift. That’s exactly why the number that matters is conditional continuation, not “how often was London green after a green Asia” in isolation.
• Uncertainty: Wilson 95% confidence intervals on every rate. Our standing rule across all studies (the same one applied in the sweep study): if the CI contains 50%, we do not call it an edge.

The results: a wall of coin flips

Take the first row seriously, because it’s the whole thesis of session-bias trading on trial. London continued Asia’s direction 49.3% of the time — 405-ish sessions out of 821 — and the confidence interval straddles 50% almost perfectly symmetrically. With n=821, this isn’t a study too small to find the effect. It’s a study large enough to say the effect isn’t there. If Asia’s direction carried even a modest signal for London — 55%, say — this sample would have caught it. It didn’t, because on this data it doesn’t exist.

The NY row reads the same: 51.6% [48.2–55.0]. The interval contains 50%. Nothing.

The “strong Asia” rescue attempt — also no

The usual counter is: raw direction is too crude — filter for conviction. A STRONG Asia session predicts follow-through.

We tested that too. Conditioning on strong Asia sessions, a tilt on NY does appear — around 55.7%. And we’re publishing it with a warning label rather than a headline, because it is fragile: the filter cuts the sample down hard, the confidence interval is wide, and a number in the mid-50s on a shrunken subsample is exactly the shape of result that evaporates out-of-sample. It is a lean worth monitoring in future data. It is not an edge, and we won’t dress it as one.

Notice what just happened, because it’s the quiet lesson of this study: the tighter you make the filter, the better the number looks and the less it means. That ratchet — slice until something clears 55%, publish the slice — is how most session “statistics” you’ll encounter get manufactured. The CI and the n are the antidote, which is precisely why sellers of session bias never show you either.

The 74.8% trap: when a real number lies

Now the interesting one. NY matches London’s direction 74.8% of the time [71.7–77.6]. The CI is far from 50%, n=821, rock solid. So sessions do predict each other after all?

No — and this row is in the table as a worked example of how a statistically bulletproof number can still mislead. The NY and London sessions overlap. For a chunk of the afternoon, “London” and “NY” are literally the same hours of the same tape, counted in both sessions. Two measurements that share their data agree with each other — that’s not prediction, that’s double-counting. A correlation between a session and itself will always look magnificent.

So the 74.8% is largely an overlap artifact. It tells you almost nothing about whether the London morning predicts the NY afternoon — the question a trader actually cares about — because the shared hours do the heavy lifting in the statistic. Anyone who quotes session-agreement numbers on overlapping sessions without flagging this is either unaware of it or hoping you are.

This is the single most transferable lesson in the study: a tight confidence interval validates the measurement, not the interpretation. The 74.8% is measured correctly. It just doesn’t mean what it appears to mean.

Why publish a null result?

Because a verified “no” changes behavior, and most of what this industry publishes never does.

A null result deletes an input. If Asia’s direction carries no information about London — and at n=821 with a CI of [45.9–52.7], it carries none — then every decision it was feeding was noise-driven: the pre-London bias, the “overnight tone” journal column, the skipped setup because it fought the Asia candle. Dropping a noise input simplifies the process for free. That’s rare and valuable.

It also tells you what Asia is actually for. Asia’s direction predicts nothing — but Asia’s structure still matters, just through an entirely different mechanism. The Asia session builds the overnight range, and its extremes become reference liquidity for the day. When we measured what happens after those levels get swept, we found the only clean fade edge in our entire sweep study: the Asia Low, swept in a Transition regime, closed back above the level by end of day 65% of the time [53–76] (n=63) — the full breakdown is in the liquidity sweep study. Asia matters as terrain, not as forecast. Direction is the wrong feature; levels are the right one.

And for where the actual action concentrates once London hands off — the NY block and its ~23% of daily volume — see the volume-by-hour numbers.

There’s a business reason we publish these too, and we’ll say it plainly: our indicators display historical probabilities on the chart, so our research has to be the kind you can check — including when the answer is “no edge.” A pipeline that only ever finds tradeable patterns isn’t a research pipeline; it’s a marketing department. This study, the FOMC study’s “no directional edge” conclusion, and the OPEX round-number theory we tested and rejected are the proof of process. If the sessions themselves — boxes, ranges, sweep tracking, regime context — are what you want on your chart, that’s what SLR (Session Liquidity Regime) does — and SLR Lite (sessions + regime, without the liquidity block) is free.

Limits

• Direction is a blunt instrument. Session direction (up/down over the session) is the folklore’s own claim, so it’s what we tested — but it compresses a whole session into one bit. Range, structure, and level interactions are separate, richer questions; some of them (sweeps of Asia levels) do carry conditional signal.
• Close-based observation, not a backtest. These are directional frequencies, not simulated trades — no entries, stops, or costs anywhere in the numbers.
• The 55.7% tilt might firm up — or vanish. Fragile means fragile in both directions. We’ll re-run it as the sample grows; until the CI clears 50%, it stays a footnote.
• One instrument, one window. XAU/USD spot, 2023–2026, with a 53.2% Asia-up drift baked into the era. A different macro regime could shift the baselines. Historical frequencies are not guarantees.
• Session definitions matter. Different session clock boundaries would perturb the numbers; none of the perturbations we’d consider plausible turns 49.3% into an edge.

Bottom line

Across 821 measured sessions, London continued Asia’s direction 49.3% of the time [45.9–52.7], and NY 51.6% [48.2–55.0] — the “Asia sets the tone” theory is a coin flip on Gold, full stop. The strong-Asia tilt (~55.7%) is too fragile to trade, and the impressive-looking 74.8% NY–London agreement is largely an overlap artifact, not a forecast.

Asia still earns its place on your chart — as the builder of the levels the day will hunt, not as a crystal ball. Knowing the difference is worth more than most “session bias” content ever sold.

Delta Terminal ResearchStudies run on 3.5 years of XAU/USD spot data, developed hand-in-hand with a full-time Gold day trader. We publish the negative results too.