Research · test #3
Is Uptober real? No edge
"Uptober" - the idea that bitcoin tends to rise in October - comes back every autumn. The past numbers do look good. We tested whether they are better than what chance produces with this little history. They are not, and no other month is either.
In short: October averaged +19.8% (median +14.6%) and closed higher in 10 of 13 years (2013-2025). Shuffling the month labels at random produces an October at least this good 12.5% of the time (p = 0.125). After correcting for testing all 12 months, no month stands out. Live table: bitcoin monthly returns.
What we already knew, and what we tested
That October has often been green is common knowledge; we did not discover it. The question is different: with 13 Octobers, could a "best month" this strong simply be luck? The rules were written down before loading the data:
- Data: Bitstamp BTC/USD daily candles; a month's return is the last close of the month over the last close of the previous month. 165 complete months, January 2013 - September 2026.
- Test: shuffle which return belongs to which month 100,000 times and count how often the shuffled "October" averages at least as much as the real one.
- Edge requires: p below 0.0042 (0.05 split across 12 months), October positive in at least 75% of years, and October beating the other months in both 2013-2019 and 2020-2025.
Result
| Month | Average | Median | Up years | p (chance) |
|---|---|---|---|---|
| October | +19.8% | +14.6% | 10 / 13 | 0.125 |
| November | +41.3% | +8.8% | 8 / 13 | 0.035 |
| February | +11.1% | +11.6% | 10 / 14 | 0.31 |
| September | -2.4% | -2.3% | 6 / 14 | 0.056 (low side) |
| June | -2.1% | -0.6% | 7 / 14 | 0.061 (low side) |
October did beat the other months in both halves of the history (2013-19: +21.2% vs +14.1%; 2020-25: +18.2% vs +3.6%) - but p = 0.125 is far from the threshold. November's p = 0.035 looks better until you remember two things: with 12 months tested, one of them landing below 0.05 by luck is expected, and November's average comes almost entirely from +450% in November 2013 (its median is +8.8%).
October, year by year
| 2013 | +61.2% | 2018 | -4.5% | 2022 | +5.5% |
| 2014 | -13.6% | 2019 | +10.3% | 2023 | +28.5% |
| 2015 | +31.3% | 2020 | +28.2% | 2024 | +10.9% |
| 2016 | +14.6% | 2021 | +40.0% | 2025 | -4.0% |
| 2017 | +48.7% |
Would an "October only" strategy have paid?
Holding bitcoin only through the 13 Octobers multiplied money 8.5 times. Holding through the whole period multiplied it about 6,300 times. Even if October really were special, sitting out the other eleven months would have missed almost all of bitcoin's growth.
What this test does not say
- "No edge" here means not distinguishable from chance with 13 years of data, not "proven random". A moderate seasonal effect could exist and still fail this test.
- 2013-2017 was a tiny market with different participants; months are not independent (bull and bear cycles span many of them).
- One exchange's prices (Bitstamp). Other exchanges differ slightly at month-ends.
Method and other tests: research. Not investment advice.
FAQ
Is Uptober real?
In the past data, yes: October was the best month by median and positive in 10 of 13 years. As a statistical edge, no: with 13 Octobers, a month this strong appears by chance about one time in eight.
Is September bad for bitcoin?
It averaged -2.4% (p = 0.056), just short of significance, and the last four Septembers closed higher.
Would holding bitcoin only in October have worked?
It would have multiplied money 8.5 times versus about 6,300 times for holding throughout.