12 March 2026
Why average correlation hides the week that matters
A calm five-year average can flatter a book that fails in the same three weeks every crisis.
Ask for “the correlation” between two equity benchmarks and you will often receive a single number: the average pairwise figure over five or ten years. That number is tidy. It is also frequently useless for the decision in front of a risk committee.
What an average erases
During long quiet stretches, major equity indices often drift with only moderate linkage. Those months dominate a long sample. The handful of weeks when liquidity thins and indices lurch together barely move the average — yet those are the weeks that decide whether a geographic sleeve actually cushions the book.
A better habit
When we prepare global index correlation analysis, we insist on naming at least one calm window and one stress window alongside any rolling series. The comparison is the point. If the calm figure is 0.35 and the stress figure is 0.82, the committee learns something the average will never say.
Practical takeaway
Before your next policy review, list the crisis periods your trustees still remember. Compute relationships there first. Then look at the long average — as context, not as the headline.