Measured on 49 perpetuals and 4,565 signals: what a 4-hour breakout really adds depending on whether the daily and the weekly confirm — and why context weighs more than the trigger.
A 4-hour breakout cannot be judged on its own. It is the most widely repeated idea in multi-timeframe analysis, and it is almost always asserted without a number. It can be measured — and the result shifts the question.
The previous articles established two things: the EMAs 13, 25 and 32 give a mechanical definition of trend, and the Stochastic RSI says nothing about direction, only about timing. What remains is the question the screener's four timeframes pose together: when a short signal appears, what are the higher scales worth in the decision?
To compare timeframes, the same rule has to apply to each. Otherwise you are comparing one method with another, not one scale with another.
The definition used is the one from the first article: an asset is in a trend when its close is above the EMAs 13, 25 and 32, and those three averages are stacked in the order 13 > 25 > 32. It applies identically in 4 hours, daily and weekly.
The signal is the entry into that state in 4 hours: the previous candle was not in it, this one is. The context is the state of the two higher scales at that instant — reading only candles that have already closed, so that nothing is borrowed from the future.
Four possible contexts, then: both confirm, daily only, weekly only, neither.
Measured on 49 Bybit perpetuals among the most traded, from 1 May 2025 to 12 September 2026 — 3,000 four-hour candles per contract, roughly 500 days. 4,565 signals recorded.
Each signal is compared with a control: the average return of any candle taken in the same context. Without that control, one would credit the signal with what belongs to the market.
First at one day:
| Context at the time of the signal | Signals | Signal | Control |
|---|---|---|---|
| Daily and weekly both confirm | 532 | +0.62 % | +0.14 % |
| Daily only | 1,184 | +0.32 % | +0.33 % |
| Weekly only | 157 | +0.66 % | +0.24 % |
| Neither | 2,692 | −0.21 % | −0.01 % |
At this horizon the signal does add something — but only where a higher scale confirms: it adds half a point when both are favourable, as much when the weekly alone is. Where nothing confirms, it subtracts: −0.21 % against a control at break-even. A breakout with no backing does worse than doing nothing in the same conditions.
| Context at the time of the signal | Signal | Control | Gap |
|---|---|---|---|
| Daily and weekly both confirm | +2.23 % | +2.26 % | −0.03 |
| Daily only | +0.75 % | +0.81 % | −0.06 |
| Weekly only | −0.77 % | −0.01 % | −0.76 |
| Neither | −0.31 % | +0.09 % | −0.40 |
The edge has gone. At seven days, the gap to the control is zero or negative in all four contexts. What the signal contributed was consumed within a day, and beyond that it distinguishes nothing.
That is useful information in itself: a 4-hour trigger has a shelf life of about a day. Using it to justify a position held for a week means relying on a stale measurement.
The clearest result of this measurement is not in the “signal” column. It is in the control column.
At seven days, any candle taken while the daily and the weekly are favourable returns +2.26 %. That same ordinary candle, when neither is, returns +0.09 %. More than two points of difference — obtained with no signal at all, purely by being in the right context.
No line in the table shows a trigger capable of approaching that gap. The conclusion follows: choosing the context is worth more than refining the trigger. That is the exact opposite of how one instinctively works, hunting for the setting that will do better.
The finding echoes the previous article, obtained on a completely different instrument: the share of winning signals barely moves — 46.9 %, 42.4 %, 49.0 %, 43.0 % depending on context, against 45 to 50 % for the controls. In three cases out of four, the signal lowers the frequency of correct decisions. What changes from one context to another is the amplitude of what follows, never the probability of being right.
One more figure the returns do not show: how the signals are distributed.
That is a mechanical property, not chance: a short scale produces signals constantly, including — especially — during bounces inside a bearish structure. Most of what an eye fixed on the 4-hour chart perceives as opportunities belongs to the least favourable category.
The job of a filter is therefore not to find more signals. It is to discard most of them.
The selector at the top of the table offers four timeframes: 4 hours, 1 day, 3 days, 1 week. The moving-average dots recalculate entirely across them.
The reading this measurement suggests is to go through them from the longest to the shortest, not the other way round:
The 3-day timeframe, absent from this measurement, sits between the daily and the weekly: it mainly serves to break the tie when those two contradict each other.
The “weekly only” context rests on 157 signals. That is few. Its swings from one horizon to the next — from +0.39 to −0.76 of a point — should not be interpreted: the sample does not support it. The other three contexts, between 500 and 2,700 cases, are solid.
The window covers sixteen months. It carries the mark of that period's conditions. The measurement establishes what happened on this sample, not a regularity verified across several cycles.
Averages hide highly asymmetric distributions. As everywhere on this market, a few large-amplitude moves carry the positive figures. “+2.26 % on average” does not translate into “I will make money”.
Only one trigger was tested. Entry into a trend by the EMAs 13/25/32 is not every possible signal; another trigger would produce other numbers. What should hold, on the other hand, is the hierarchy: context decides, the trigger dates.
What remains fits in one sentence, and it is more demanding than it sounds: the useful work is not spotting signals, but knowing in which context you refuse to look at them.
This is not investment advice. This article sets out technical-analysis concepts for educational purposes. Past performance is no guide to future performance, and trading crypto-assets carries a risk of losing the entire capital committed. Legal information.