# Do Extreme Readings Predict the Next Week?

Every dashboard in crypto colors its gauges: extreme fear in green because "it is a buying opportunity", scorching funding in red because "a flush is coming". This note checks five of the most-quoted gauges against our recorded daily history and asks one narrow question: conditioned on today's band, does the distribution of BTC's next-7-day return actually differ from any other day's?

Source: https://bykaranteli.com/research/band-conditioned-returns · attribute to ByKaranteli and quote the timestamp.

Published First published 2026-08-13 · tables recompute daily from the archive.

## The question

Take a gauge's full recorded daily history and split it into quantile bands (five where history is deep, three where it is short). For each band, look at every day that fell in it and measure BTC's return over the following 1, 3 and 7 days. If the folklore is right, extreme bands should show forward distributions that clearly separate from the all-days base rate.

This is deliberately narrow. It does not test strategies, entries, exits or costs. It asks only whether the conditional distribution moves.

## Method

Data: one value per UTC day per gauge from our metric archive, sources dated (live-recorded or backfilled from dated public series). Forward returns join against recorded BTC daily closes at +1, +3 and +7 calendar days.

Bands are quantiles computed over the full available history at read time, so no band boundary is hand-picked. Bands with fewer than 30 days are suppressed rather than shown thin. The median and the share of positive outcomes are reported per band, with the all-days base rate on every table.

The tables below are the live product of that method: they recompute from the archive as it grows, so this note can change its own conclusions. That is intended. A research note that cannot be moved by new data is marketing.

## Results, gauge by gauge

Each table is live. "Today" marks the current band. Read the 7-day column against the base rate at the bottom of each table before believing any narrative about the band you are in.

- coinbase_premium_pct: /md/factors/coinbase_premium_pct
- fear_greed: /md/factors/fear_greed
- dvol_btc: /md/factors/dvol_btc
- funding_btc_daily_pct: /md/factors/funding_btc_daily_pct
- stablecoin_total_mcap_busd: /md/factors/stablecoin_total_mcap_busd

## What this does not say

No costs, no execution, no strategy: a band whose median 7-day move is +1% is not a trade, it is a distribution. Median differences smaller than typical daily noise are reported because hiding them would be dishonest, not because they are actionable.

Bands are computed over each gauge's own history, which mixes regimes: a "very high" stablecoin float band is dominated by recent years. The tables state n and the record span for exactly this reason.

And the standing caveat of all conditional history: distributions describe what happened after similar readings, not what must happen next. Correlation is not causation, and eight years is one market's lifetime, not a law of nature.

## Data and reproduction

Everything in this note derives from the metric-daily dataset; the per-gauge tables come from the public context API with a free account key. Recompute any table yourself:

## Related

- https://bykaranteli.com/research
- https://bykaranteli.com/factors

Everything in one fetch: https://bykaranteli.com/llms-full.txt