Move from the sample to the mechanism
A single estimate compresses multiple environments into one number. Start by comparing calm and stressed periods, shorter and longer windows, and episodes with different liquidity conditions.
Disagreement across windows is useful evidence. It says the portfolio role is conditional and should be expressed as a range of plausible behaviours.
Two assets may move together because the same investors de-risk, because funding tightens, or because a broad macro narrative dominates. Those channels have different persistence.
Pair every correlation exhibit with a mechanism ledger: participant base, funding route, market depth, and likely exit path. The ledger turns a chart into a research hypothesis.
- Compare rolling windows
- Annotate stress episodes
- List mechanisms that could break the relationship
Look directly at the downside
Average co-movement can obscure the periods that matter most to a portfolio. Conditional checks around sharp drawdowns, volatility jumps, and thinner liquidity expose that weakness.
When diversification fades under pressure, the decision shifts toward loss budgets, rebalance capacity, and implementation.
Write a conditional conclusion
The useful output is not a permanent category. It is a statement of the environments in which an asset may add an independent return source, and the environments in which it may behave like a high-beta exposure.
That conclusion should end with monitoring triggers. Correlation research stays alive only when its assumptions can be revisited.