Rolling Correlation

Why a single correlation number is rarely enough.

What “Rolling” Means

A rolling correlation calculates correlation repeatedly over a moving window (for example, the last 30 days), producing a time series. This reveals how relationships change.

What Rolling Correlation Shows Well

• Regime shifts (from “everything moves together” to “dispersion”)
• Narrative cycles (sector rotation, memecoin episodes, L2 rotations)
• Liquidity-driven phases (correlation spikes during stress)

Common Window Lengths

• 7–14 days: fast, noisy
• 30 days: a common balance
• 90 days: slower, smoother, can lag turning points

Two Common Mistakes

• Over-reading short windows: a week can be dominated by a single event.
• Ignoring volatility: correlation and volatility interact; stable periods can produce low correlations that vanish in stress.