When a single correlation number misleads a committee
Why averaging co-movement across calm and stress years can hide the breaks that matter for policy language.
Investment policy appendices still love a tidy figure: “equity–equity correlation of 0.7.” The number is rarely false for the window that produced it. The trouble is the window.
When calm years dominate the sample, the average can sit high while the episodes that hurt the mandate — liquidity squeezes, rate path surprises, regional selloffs — show different meshes entirely. Committees then write caveats that sound prudent but point at the wrong relationship.
In our briefings we separate rolling views from regime slices and ask the chair which episodes deserve a named place in the minutes. Often the useful output is not a replacement average, but a short list of breaks with dates attached.
If your room still cites one figure from a consultant pack, ask which years drove it — and what the same pairs did in the last two stress windows you actually lived through.