Autocorrelation
Does this week's return tell you anything about the direction of next week's return?
In plain words
Autocorrelation measures how closely a series of returns resembles its own past. Applied to returns, it asks whether a rise tends to be followed by another rise, or a fall by another fall. A reading near zero means the pattern carries no memory; positive values mean same-direction clustering, negative values mean reversals.
Why it matters
Many strategies quietly assume momentum, that strength persists. Autocorrelation near zero says that assumption earns nothing on direction, even though volatility still clusters, quiet weeks followed by wild ones. Sizing, entry timing and holding periods built on expected follow-through fail when the next move is independent of the last one.
In a real week: BTC, week of 17 June 2019
In the week of 17 June 2019, BTC rose 21.81%, closing at 10906.07, in the 97th percentile among 70 weeks of history. The four prior weeks returned +5.73%, +1.32%, -12.58% and +17.38%: three up and one down. That mixed sequence gave little reason to expect the extreme result that followed, illustrating how weakly one period's move predicts the next.
A common mistake
Assuming a strong week predicts another strong one. Returns usually show little autocorrelation on direction, and momentum measured at one interval may vanish at another.