The same 20 % gap is worth 1.7 standard deviations on one asset and 15.7 on another. Measured across 45 perpetuals and 28,229 observations: neither the percentage nor its normalised version predicts direction — and the spectacular figure suggesting otherwise came from nine candles.
"That asset is too stretched." The phrase is everywhere, and it is almost always said without a unit. Stretched by how much, measured how, and compared against what?
The previous articles set out a definition of trend, the role of the long moving averages as structure, and showed that the Stochastic RSI shifts the size of a move without changing the odds of being right. Stretch raises a neighbouring but distinct question: when price pulls away from its moving average, does that gap announce anything?
The gap is naturally measured in percent: price is 20 % above its EMA 200. That is convenient, and it is misleading, because 20 % does not describe the same event from one contract to the next.
Across the 45 contracts measured here, median daily volatility runs from 1.28 % per day on XAUTUSDT to 11.84 % per day on AKEUSDT. Set against each asset's own habitual movement, the same 20 % gap is worth:
| Contract | Median volatility | What a +20 % gap is worth |
|---|---|---|
| AKEUSDT | 11.84 % / day | 1.7 standard deviations |
| FARTCOINUSDT | 6.79 % / day | 2.9 standard deviations |
| LTCUSDT | 3.14 % / day | 6.4 standard deviations |
| BTCUSDT | 2.14 % / day | 9.4 standard deviations |
| XAUTUSDT | 1.28 % / day | 15.7 standard deviations |
From one contract to another, the same figure covers situations that differ by a factor of 9.3. On AKEUSDT, 20 % above the average is an ordinary week. On XAUTUSDT it is an event the history barely contains. Filing both under "stretched by 20 %" is comparing metres with feet and skipping the conversion.
The correction is immediate: divide the gap by the asset's own volatility. What comes out is no longer a percentage but a number of daily standard deviations — how many normal days of movement separate price from its average. That unit is comparable across contracts. Whether it predicts anything is another matter.
Measured on the 45 most traded Bybit perpetuals with sufficient history, on daily candles: 28,229 observations, roughly 627 days per contract.
For every closed candle, three quantities and one outcome:
Fifteen contracts among the sixty most traded were dropped for lack of history: an EMA 200 needs 200 candles, volatility another 30, and a 20-day forward return 20 more. That constraint deserves an article of its own.
No stretch figure means anything without knowing what an ordinary candle from the same sample returns. That control holds a surprise:
| Horizon | Mean return | Median return | Share rising |
|---|---|---|---|
| 5 days | +0.46 % | −0.69 % | 46.3 % |
| 10 days | +1.06 % | −1.13 % | 46.3 % |
| 20 days | +2.68 % | −2.56 % | 43.7 % |
The mean is clearly positive, the median clearly negative, and fewer than one candle in two is followed by a rise. Both figures describe the same sample and tell opposite stories. The mean is dragged upward by a minority of enormous moves; the typical case loses.
This is the defining property of the universe, and it makes the average forward return dangerous to handle on its own. Everything below is measured against it.
Sorting the 28,229 observations into deciles, first on the raw gap and then on the normalised gap, the spread between the first and the last decile at 20 days is 8.08 points for the raw gap against 6.67 points for the normalised one. The percentage — the incomparable unit — separates better than the comparable one. The starting hypothesis is wrong.
But it is the share of rising cases that actually settles the question:
| Situation at 20 days | Observations | Share rising | Median |
|---|---|---|---|
| Far below the average (≤ −50 %) | 2,251 | 47.4 % | −1.26 % |
| Close to the average (−10 to +10 %) | 5,660 | 43.3 % | −2.52 % |
| Far above the average (≥ +80 %) | 723 | 47.3 % | −1.99 % |
| Beyond 2 standard deviations | 23,473 | 44.1 % | −2.39 % |
| Control | 28,229 | 43.7 % | −2.56 % |
Four points of spread at most, and not in the direction anyone expects: both extremes rise slightly more often than the middle, and the medians stay negative throughout. An asset far above its average does not come back down more often than any other. A collapsed asset does not bounce more often either.
The averages seemed to say otherwise. The most stretched decile above the average shows +3.88 % at 20 days against +2.68 % for the control, and gaps beyond +80 % reach +12.52 %. Enough to believe in a signal.
Two checks dissolve it.
Concentration. Those 723 observations beyond +80 % come from only 21 contracts, and three of them supply 39 % of the group. Give every contract equal weight rather than every candle, and the +12.52 % becomes −1.20 %. The result did not shrink: it changed sign.
The per-contract detail. Inside the most stretched decile above the average, 25 contracts out of 40 have a negative mean return at 20 days; the median across contracts is −3.40 %. The positive pooled average rested on a handful.
The clearest case sits at the other end. The decile stretched furthest below the average showed +75 % under equal weighting — a spectacular number. It comes from a single contract, AKEUSDT, appearing nine times in that decile with a mean return of +1,976 % at 20 days. Nine candles from one asset produced the entire result.
This is not an outlier to be cleaned away: that rebound genuinely happened. It is a matter of not presenting an episode as a regularity.
The screener's stretch column reports how far price sits from its moving averages. In the light of the above, it reads as a measure of position, not a signal: it says where price stands relative to the asset's own habits, not what it will do next.
That reading stays useful, for at least two reasons. It shows the distance between price and the level where the moving average might support it, which bears on position sizing and stop placement. And once normalised by volatility, it finally allows two contracts to be compared — which the raw percentage forbids.
It says nothing about timing. A gap usually closes in the end, but by which mechanism and over what period — price falling back, or the average rising to meet it — is a separate question, and it is the subject of the next article. That one will reuse the normalised unit set out here.
It says nothing about horizontal levels either, nor about what happens when a moving average coincides with an old floor. And it covers the most traded contracts holding 250 candles: recent listings, the majority on Bybit, are absent by construction.
One finding carries beyond this subject. In this universe, the mean and the median of the same sample point in opposite directions. Any measurement showing only an average should be treated as incomplete — including, and especially, when it gives the answer you were hoping for.
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.