Correlation is a moving relationship. A single full-history number often describes no regime that investors actually experienced.
A 30-session correlation reacts quickly but can be noisy. A 90-session measure changes more slowly and provides regime context. Viewing both shows whether a new relationship is emerging or whether a short burst sits inside a stable longer pattern.
Two assets can move in the same direction while one materially outperforms the other. A useful relationship read therefore includes normalized paths, rolling correlation, relative return, and drawdown. Each answers a different question.
Testing many lags almost guarantees that one looks impressive. The relevant result must be stable across nearby periods and repeated in earlier windows. A lead-lag label is descriptive evidence about timing, not proof that one market causes another.
Rather than matching each asset separately, compare the current joint profile with prior windows: short- and long-window correlation, relative leadership, agreement, and path shape. If no candidate clears a stated credibility floor, showing no match is more useful than forcing one.