Do EUR/USD Level Clusters Actually Predict Follow-Through? 1,462 Breaks Say Yes — Modestly
Followed through (%)
"Levels matter" is trading canon, but most chart commentary treats every support or resistance line the same way — a round number gets drawn with the same weight as a prior session open or an old swing point. We tested whether that's actually true: does a level predict anything on its own, or does it only matter when several independent signals stack up at the same price?
We graded every meaningful EUR/USD level break over five months by how many independent factors converged on it — a round number, a prior session open or close, a swing high or low, and similar recurring reference points — then tracked whether price followed through or reversed. 1,462 graded breaks.
Confluence beats a lone line, roughly two to one
| Grade | What it means | Follow-through rate |
|---|---|---|
| A | 3+ independent signals converge in a tight band | ~62–66% |
| B | 2 signals converge | inconsistent month to month — not tradeable |
| C | a single, isolated signal | ~45–51% |
A genuine cluster of independent signals is meaningfully predictive. A lone line by itself is close to a coin flip, and if anything slightly worse. This is the most robust single finding in the data.
It holds up month after month, in every regime
Trending months, ranging months, quiet months — the A-grade rate barely moves:
The rate stayed in a tight 58–68% band across all five months, and the same cluster grade worked about as well in every trading session we checked. That stability matters more than the headline number: it is a modest, persistent bias, not an edge that flares up once and disappears.
What doesn't hold up: the popular "confirmations"
Three signals traders commonly use to confirm a level break did not survive a bigger sample:
- Volume expansion. A volume spike on the break candle looked like a strong confirmation in an early read, but barely replicated once the sample grew — a few extra points on the strongest grade, and it actively hurt the weaker grades.
- Session timing. An early pass suggested certain sessions were more reliable than others. At scale, that map fell apart: the strongest cluster grade works in essentially every session, and the weakest grade fails in every session. It's the level, not the clock.
- Higher-timeframe trend alignment. Weak on its own, and it inverted during counter-trend stretches — unreliable enough to discard as a filter.
One promising thread, with an honest caveat
Whether there was open room to the next graded level ahead of the break looked close to automatic in a small early sample, but softened considerably as the sample grew — still a useful directional tilt, not a guarantee, and it couldn't even be tested in months where levels sat too close together to leave open road. Waiting for a retest or reclaim of the level, rather than reacting to the first touch, showed a similar promising-but-unproven pattern. Both are worth watching, neither is confirmed yet.
What this means
- Weight your levels instead of drawing them all the same: a band where three or more independent signals stack up is real; a single isolated line mostly isn't.
- Don't lean on a volume spike or the trading session as your confirmation — in this sample, the cluster itself carried almost all of the signal.
- Treat this as a modest, persistent bias (~62–66%), not a high-win-rate system. It changes where you look, not how big you should size.
Studies like this become the filters inside a documented playbook — the research → playbook → backtest → live loop, locked to one instrument at a time, rather than a scanner firing on everything.