Every August a chart circulates showing Bitcoin’s average return by calendar month, and every August someone builds an argument on it. The chart is real. The sample behind each bar is around a dozen observations, which is not enough to distinguish a pattern from noise.
This article is for informational and educational purposes only. It is not financial advice, investment advice, or a recommendation to buy, sell, or hold any security, cryptocurrency, or financial product. Always verify data with official sources before making financial decisions.
The monthly return record
Short answer: Bitcoin has a usable price history from roughly 2013, giving about thirteen observations per calendar month. August’s average has been weak-to-mixed depending on the start date and whether the average is a mean or a median. The dispersion around that average is enormous — individual Augusts range from large gains to large losses — which is the part the chart does not show.
What thirteen observations can and cannot support
This is the central point and it is worth being precise about.
With thirteen observations and monthly returns that routinely swing 20% or more in either direction, the standard error on a monthly mean is very large — large enough that almost any calendar month’s average is statistically indistinguishable from zero and from every other month’s average. A difference between August’s average and October’s average that looks dramatic on a bar chart typically fails any test of significance.
There is a further problem. Testing twelve months for a pattern means running twelve tests, and at conventional significance thresholds you would expect roughly one to appear “significant” by chance alone. Calendar-effect research across asset classes has repeatedly found that patterns identified this way fail out of sample — which is the fate of most published seasonal anomalies in equities as well.
Midterm years: three observations, one weak average
A refinement sometimes offered is to condition on the US electoral cycle, comparing August returns in midterm years specifically. That reduces the sample from thirteen to three.
Three observations cannot support an average, a confidence interval or an inference. Presenting a mean of three numbers as a seasonal tendency is a presentation choice, not an analysis. This applies equally to halving-cycle conditioning, which faces the same arithmetic — Bitcoin has had a small number of halvings and each occurred under different market conditions.
Liquidity, holidays, and the mechanical explanations
The explanations offered for summer weakness are drawn from equity market folklore: reduced participation during northern hemisphere holidays, thinner liquidity, fewer institutional decisions.
These translate poorly to crypto. The market trades continuously across every time zone, participation is globally distributed rather than concentrated in one hemisphere, and there is no equivalent of a corporate calendar or an earnings season structuring activity. Explanations imported from equity seasonality assume institutional structures that crypto largely lacks.
Where a mechanical explanation does have some basis is in derivatives expiry clustering and the timing of large scheduled unlocks, both of which follow calendar patterns. Those are specific, datable events rather than a diffuse seasonal tendency.
Seasonality as context, not as a trade
There is a defensible use for this data, and it is narrower than how it is usually deployed.
Knowing that a given month has historically been volatile, or that returns have been widely dispersed, is useful context for position sizing and expectations. It is a statement about the distribution of past outcomes rather than a directional forecast, and it does not require the sample to be large enough for significance testing.
What the data cannot support is a trading rule. Any strategy of the form “reduce exposure in August” rests on an average of thirteen numbers with a standard error wide enough to encompass the opposite conclusion.
Mini glossary
- Seasonality. A recurring pattern tied to the calendar rather than to underlying conditions.
- Sample size. The number of independent observations available. Determines how much confidence any average can carry.
- Multiple comparisons problem. The tendency to find apparently significant results by chance when testing many hypotheses.
- Out-of-sample testing. Evaluating a pattern on data not used to identify it. Most calendar effects fail this.
What this article does not conclude
Nothing here says August will be weak, strong or anything else. The argument is about what thirteen observations can support, and the answer is: very little, regardless of which direction the average points.
Monthly return data is available from several public sources and calculations differ depending on whether returns are measured close to close, on which exchange, and from which start date. Charts from different providers do not agree.