Rolling correlation recomputes the correlation between two assets repeatedly over a sliding window, producing a line that shows how their relationship has evolved. Instead of one figure for the whole period, it reveals whether two markets have been consistently linked or have drifted apart and back together.
For a trader, rolling correlation checks the health of a relationship. A single matrix cell might say gold and the dollar are inversely linked on average, but the rolling view shows whether that link held steady or broke down recently, which matters before relying on it.
The window length is a smoothing choice, not an accuracy setting. A short window reacts quickly but amplifies noise, while a long window lags real change, so the chosen lookback belongs in any read. Strengthened or weakened relationships prompt questions about drivers and regime rather than forecasts.