Research library

When London (07:00-09:00 UTC) breaks gold's Asian range, price is back inside the range at the 09:00 checkpoint on 60.8% of sessions (n=79). The initial poke past the level is not the discriminator — its median is 2.58 points on the breaks that fail and 2.59 points on the breaks that hold. The separation is in maximum extension: breaks that never got more than 5 points beyond the level ended back inside 97% of the time, while every break that extended past 20 points held.

When gold's Asian session (00:00-07:00 UTC) closes in the top third of its own range, the 07:00-09:00 UTC London window breaks the Asian high first on 97.0% of sessions that break (n=33). When Asia closes in the bottom third, London breaks the Asian low first instead, on 100% of sessions that break (n=31). The tell calls which level goes first, not whether the move holds — a separate question the video on this page answers candle by candle.

The confluence at a level tells you the level is important. It does not tell you which way price will go. In this sample, that direction came from the live two to three day thrust: with-thrust breaks worked roughly 85% of the time and counter-thrust breaks roughly 13%. The single most expensive habit in the data was fading a multi-signal cluster while the trend was still running, because in a live thrust that cluster is usually a coil that price shakes through and then expands out of.

A EUR/USD level only predicts follow-through when it is a real cluster: a band where three or more independent technical signals (round numbers, session levels, swing points, and similar) converge. Those breaks follow through ~62-66% of the time, stable across every month and every session. A single, isolated line resolves worse than a coin flip (~45-51%). Volume spikes and session timing, the confirmations most traders lean on, added almost nothing once the sample grew.

EURUSD range boundaries resolve as breakouts, not reversions, by roughly 6-to-1 (85.4% vs 14.6% of resolved touches, n=3,111). The bias holds across almost every session and range-maturity bucket we tested — reversion is the exception you have to filter hard for, not the default expectation at a boundary touch.

S&P 500 opening-range breakouts (first 60 minutes of the session) resolve very differently by session. London (07:00 UTC open, n=131) round-trips — breaks one side, then the other — 53.4% of the time, more than triple New York's (13:00 UTC open, n=130) 16.2%. Clean single-direction success is similar in both (19.1% London vs 20.8% NY); the difference is entirely in how often the failed side gets swept too.

S&P 500 opening-hour direction predicts the cash session's close direction 70.5% of the time overall (n=606), climbing to 81.5% when the opening hour is wide (>1.5× H4 ATR, n=54) versus roughly a coin flip (63.6%) when it's narrow. A directional opening hour also tends to keep extending: after an up open, the opening high breaks again within four hours 81.0% of the time versus only 35.2% for a reversal back through the opening low.

A EURUSD pullback in a confirmed trend continues just 48.7% of the time — the raw 'buy the dip' edge is nil. What restores it is character: shallow pullbacks continue 75% of the time vs 37% for deep ones, slow grinds continue 67% vs 28% for sharp drops, and pullbacks that bottom before touching the H1 EMA20 continue 71% vs 39% for those that reach it. Depth, speed, and timing are the signal — not the dip itself.

Elevated fear is a scale-up-long regime, not a reason to sell. Sorting 13 years of sessions by realized volatility, the calmest fifth returned +0.32% over the next 20 days while the most panicked fifth returned +2.06% at a 72% win rate — roughly 6× the forward return. Confirmed against real VIX: calm (~12) → +0.55%, panic (~28) → +2.41%. It is a sizing overlay with fat left tails, not a standalone entry.

A EURUSD breakout is only a coin-flip-plus until you condition on displacement. Breaks that push less than ~3 pips past the level fake out 60% of the time; breaks that displace hard follow through 90%+ of the time. The candle's conviction, not the level itself, is the signal.

S&P 500 gap-fill probability is a near-monotonic function of gap size. Tiny gaps (<5 pts) fill ~96% of the session and are noise to fade; large gaps (>50 pts) fill only ~20% and extend — they are signal to follow. Relative to ATR the split is even cleaner: under 0.25× ATR fills 94%, over 2× ATR fills 9%.

Gold's opening-range breakout has a resolved success rate around 32% in both London and New York — but the failure mode differs sharply. London breakouts get faded and reversed (47% sweep both sides) far more than New York breakouts (22%). The session you trade the ORB in matters more than the ORB itself.

Gold spends ~75% of hours in RANGE and only ~17% in TREND, and those trends are brief (median run of a couple of hours) while ranges persist. London sweeps the Asian range 97% of the time. Gold is a mean-reverting, liquidity-seeking instrument punctuated by short violent trends — trading it as a persistent trender fights its base rate.

Buying strength and shorting weakness with no filter is a slight loser across ~25,000 events (47.6% hit rate). The continuation edge is entirely in the context around the move: volume expansion and a pre-existing trend flip the odds to 80%+, while over-extended moves with no trend behind them mean-revert hard.

A EURUSD liquidity sweep — a wick past a swing level that closes back inside — reverses only about 30% of the time and continues ~68% of the time. Fading every sweep as a stop-hunt is a losing default. The reversal edge is a minority case you have to filter for, not assume.