Seasonality in Cryptocurrencies: Weak, Unstable, and Easy to Data-Mine
Is reported crypto seasonality stable across exchanges and sample periods?
Estimate calendar effects separately by exchange, asset, time zone and held-out period, with trading costs included. Kaiser?s empirical sample tests particular cryptocurrencies; its result cannot establish a stable seasonal edge across all venues or later regimes.
The Kaiser and Aharon-Qadan findings are bounded to their cryptocurrency samples and calendar cuts; they do not prove a durable net return edge.
Evidence map
| Aspect | Finding |
|---|---|
| What it is | The literature testing whether crypto markets show calendar effects of the kind long documented in equities: day-of-the-week patterns, month effects, turn-of-period effects. |
| Key result / formula | Kaiser tests ten cryptocurrencies for calendar patterns in returns, volatility, trading volume and a spread estimator. |
| Why it matters for backtesting | This is the reference an agent should use to slow a user down, not to encourage them. |
What it is
The accurate summary is that some patterns are reported, they are small, and they are not stable.
Key result / formula
The reported abstract specifically identifies lower average volume, volatility and spreads in January, on weekends and during summer months in that sample. It does not establish a durable net return strategy. Aharon and Qadan report Bitcoin day-of-week effects in returns and volatility using daily observations from 2010 to 2017. Those results concern their data and calendar cut; testing later periods or other venues is a separate experiment. Two structural features of the market make the question harder than in equities. There is no session and no closing auction, so a "day" is a convention of the data provider and depends on the time zone in which the bars were cut; and the market does not close, so there is no weekend gap to create the mechanism that generates several equity calendar effects. The result is a literature that reports effects whose existence depends on decisions the researcher made about how to slice continuous time.
Why it matters for backtesting
Testing a day-of-the-week effect on crypto bars is trivial to do and trivially data-mined: seven days, two directions, several currencies and a handful of sample windows generate hundreds of implicit trials, and the best of them will look significant (see [Data Snooping, P-Hacking & the Garden of Forking Paths]). The disciplined version declares the number of trials before looking, uses the same time-zone convention throughout, and deflates the result accordingly. The one robust finding to carry forward concerns volume and volatility, and leaves returns aside: activity in crypto does vary with the hour and the weekday because the humans trading it sleep, and that matters for cost and fill assumptions even where it carries no information about direction.
Source
Kaiser, "Seasonality in cryptocurrencies", Finance Research Letters 31, 2019, 232-238; Aharon & Qadan, "Bitcoin and the day-of-the-week effect", Finance Research Letters 31, 2019, 415-424. Primary source