Research · test #12

Does extreme fear mean a bitcoin bounce? Undecided

"Sentiment is extremely bearish - the short squeeze is coming." We took every time the Crypto Fear & Greed Index entered Extreme Fear since 2018, bought bitcoin and held a week, with the rules written down first. Before 2023 it did worse than a random week. Since 2023 there are too few cases to call it - so the verdict is undecided.

In short: 32 entries into Extreme Fear. A week later, after 0.2% costs: -3.44% on average in 2018-2022 (24 cases), +1.96% since 2023 (8 cases - the rule needed 15). A random week made +0.70%. A plain 10.9% weekly drop, ignoring the index, made +1.11%.

The rules (fixed before the data was loaded)

• Index: Crypto Fear & Greed Index (alternative.me), 3,185 daily values from February 2018. Signal threshold: 24 or lower. (After the run we found the index labels 25 as Extreme Fear too; with 25 included the result is the same - 34 signals, worse than random days before 2023, 10 since.)
• Signal: the day the index enters Extreme Fear after at least seven days outside it, and at least seven days after the previous signal - one signal per fear episode.
• Trade: buy bitcoin (Binance spot) at that day's close, sell seven days later; 0.2% round-trip costs, no leverage.
• Controls: every day of the period (luck), and weeks when bitcoin had already fallen as much as it typically had before a signal - to see whether the index adds anything to the price drop itself.
• Pass mark: positive average and better than random days (p < 0.05) in both 2018-2022 and 2023-2026, and better than the plain drop. At least 15 signals in each period, otherwise undecided.

Result: no bounce beyond a random week

In 2018-2022 a random week beat the fear signal 98% of the time. Since 2023 the average is positive, but with 8 cases it is well within chance (p = 0.29) - and below the 15 the rule required, which is why the verdict is undecided rather than no edge.

Every signal, one week later

Each bar is one signal, in date order (year labels: '18 = 2018). Worst weeks: 2020-03-09 (-36.8%), 2021-05-16 (-25.6%), 2018-11-16 (-22.0%). The big losses came when extreme fear arrived and the price kept falling for another week.

The index adds nothing to the price drop

Half of the Fear & Greed Index is volatility and market momentum (25% each, by alternative.me's description), so it is fair to ask whether it says anything that the price itself did not. Weeks in which bitcoin had fallen 10.9% or more - the typical drop before a fear signal - were followed by +1.11% on average (66 cases, 65% up). The fear signal did worse: -2.09%.

Other versions (not part of the verdict)

• Every Extreme Fear day, not just the first (679 days, overlapping weeks): +0.71% a week - the same as a random day (+0.70%).
• Deeper fear, index 10 or lower (16 signals): +2.94%, t 1.23 - not significant.
• Other horizons: 3 days -2.17%, 30 days -2.96% (median -2.89%).
• The mirror: shorting when the index enters 75 or higher (22 signals) averaged -0.25% a week (p = 0.32 against random days).

What this test does not say

• 32 signals over eight years is a small sample, and a few crashes (March 2020, May 2021, November 2018) weigh heavily.
• It tests one asset, bitcoin, and the plain rule "buy the first day of extreme fear, hold a week". Scaling in, waiting for the index to turn up, or other assets are not tested.
• The index methodology may have changed over the years; we used the series as published on 25 October 2026.

Data and method

Rules and code were committed before the data was loaded. Index: alternative.me public API. Prices: Binance BTCUSDT daily candles. Related: buying a leverage flush (test #11), extreme funding, and what 11 crypto backtests show. Method: methodology.

FAQ