An altcoin-versus-Bitcoin screener shows about a thousand coins, each stamped with one of three labels: uptrending, basing, downtrending. Green, grey, red. The site is well built, the data is current, and the question it answers is clear: what is strong against Bitcoin right now?
The question it does not ask: what historically followed "strong"?
We asked it. First we fixed the label's definition in writing, so we could not adjust it to the result afterwards. Then we ran 370 Binance pairs across 457,994 trading days. The result fits in one sentence, and the sentence is not "no difference".
What the label measures
A trend label against Bitcoin needs two things: a price series denominated in Bitcoin rather than dollars, and a rule. The series is the ratio of the coin's dollar price to Bitcoin's dollar price — when it rises, the coin beat Bitcoin, whatever the dollar did.
We chose the rule the way screeners usually build it: uptrending means the ratio sits above its 50-day average and the 50-day average sits above the 200-day average. Downtrending is the mirror image. Everything else is basing. Daily close, no costs, no discretion.
What matters is not whether this rule is the best one. What matters is that it was fixed before the run. Whoever tunes the parameters afterwards until the result looks good is measuring their own search, not the label. We made exactly one attempt, and the hash of the definition is in the dataset.
The floor: what any given day delivers
Before a label can prove anything, it needs a comparison. The honest comparison is not zero, but: what happened on every day, whatever label it carried?
Across all 370 pairs and every day since 2017, the median altcoin lost against Bitcoin by 2.0 % over seven days, 6.8 % over thirty and 17.3 % over ninety. Positive outcomes: 39 % of seven-day windows, 33 % of thirty-day windows, 26 % of ninety-day windows.
Put differently: in three of four quarters the median altcoin would have lost ground to Bitcoin. That is the floor. A label worth anything has to sit above it.
The label sits below it
On 8.7 % of all days a coin carried the uptrending label. What followed, in medians:
| Horizon | Uptrending | Any day | Difference |
|---|---|---|---|
| 7 days | −2.7 % | −2.0 % | −0.8 pp |
| 30 days | −9.6 % | −6.8 % | −2.7 pp |
| 90 days | −23.0 % | −17.3 % | −5.6 pp |
The uptrending label sat below the average at every horizon. Not dramatically, but consistently, and growing with the horizon.
The opposite direction is the only one with an edge: coins in the downtrending state — 58 % of all days — sat 0.4, 1.2 and 1.9 percentage points above the average. A mild reversion to the mean, not a case for buying; the median stayed negative there too.
Three checks that could have rescued the result
A result pooled over 458,000 days can mislead in several ways. We checked the three usual ones.
Overlapping windows. Counting every day counts the same ninety-day path ninety times. So we kept only every seventh, thirtieth and ninetieth day per pair. That left 5,694, 1,312 and 428 uptrending windows with no overlap. Difference: −0.8, −2.6, −3.2 points. Same sign at a hundredth of the sample.
A handful of pairs pulling everything. A pooled result can come from a few coins. So, per pair: did the uptrending median sit above that coin's own average? For 28 % of pairs at seven days, 22 % at thirty, 20 % at ninety. Four in five coins pulled the label down.
A single regime. Before 2024 the difference was −0.9, −2.9 and −6.6 points; from 2024 on, −0.8, −3.5 and −8.7. Two market phases, the same picture.
Why a clean label loses anyway
The mechanism is not a property of altcoins but of the label. A 50-day average takes weeks to move above a 200-day average. In test series with built-in trends, the uptrending stamp appeared a median of about 80 days after the trend began — never earlier than 52, never later than 115.
The instrument does see trends: in an artificial series with 600-day regimes the label delivered an edge of 4.2 points. With 120-day regimes the same label delivered −3.7 — because it turned green precisely when the reversal was due.
Translated: alt/BTC trends apparently do not usually last longer than the classifier needs to detect them. The label is not a bug in the screener. It is a description of the past being read as a description of the future.
The strongest counter-argument
Two objections come to mind, and both deserve an answer.
First: you only count coins still trading on Binance today. The dead ones are missing. True, and declared. But a dying coin spends its last months in the downtrending state. Its absence flatters the downtrending arm — not the uptrending arm. The bias works against the only edge we measured, not for the label.
Second: you use medians. A basket catches the outliers, and they make the difference. Also true — and it is why we do not quote the means. In the basing state, the seven-day mean before 2024 was +152 %; the median was −1.8 %. The gap is a few days with hundred-fold moves. Whoever defends the label with the mean is defending three individual days. And a daily equal-weight basket that catches such days earns its return from rebalancing, not from the label.
What to do with this
An alt-versus-BTC screener is an inventory, not a signal. It shows what sits at the top. The number it owes you is now here: the median altcoin loses 17 % against Bitcoin in a quarter, and the green label moves that figure down, not up.
Whoever rotates anyway should know which current they are swimming against. We had measured this once before from the other side: a rule-based rotation between Bitcoin and altcoins failed to beat holding in 45 tests — stacking more sats with altcoin rotation?. The base rate explains why that is hard.
The numbers stay alive. The dataset is recomputed monthly and available as the Knowledge Object alt_vs_btc_base_rate via the API and the Claude connector — with the definition, its hash and all three checks in the payload. If the label ever rises above the base rate, it will show there first.
The open question is the one every screener would have to answer before handing out a label: which state of an altcoin against Bitcoin — defined in advance, not found in hindsight — has ever beaten the base rate?
FAQ
What does "base rate" mean here? The median forward return of the alt/BTC ratio across all days and pairs, with no condition. It is the benchmark every label is measured against.
Why the median and not the mean? A few days with extreme price jumps shift the mean by orders of magnitude (+152 % against a median of −1.8 %). The median describes the typical day; the mean describes three outliers.
Isn't the definition of "uptrending" arbitrary? It is one of many possible ones — which is exactly why it was fixed before the run and not varied afterwards. A different definition would be a second attempt and would be counted as one.
Aren't delisted coins missing? Yes. Only pairs trading today are included. The effect flatters the downtrending arm, because dying coins end up there; it does not move the uptrending result in the label's favour.
Does the result hold for 2024 and later? Yes. Before and after 1 January 2024 the uptrending label sat below the average at all three horizons.
Where do I find the current numbers?
In the Knowledge Object alt_vs_btc_base_rate (/api/v1/knowledge/alt_vs_btc_base_rate/GLOBAL), recomputed monthly, free tier.
Not investment advice, not a recommendation, not a forecast. Historical patterns are no promise for the future.
Study the Past — Improve your Future 🥋