Start with one example. Gold pays you nothing to hold it. A government bond does. So when the “real” yield on that bond goes up, holding gold costs you more in missed income, and gold usually falls. The two normally move in opposite directions. That link isn’t a superstition — it comes from how each one actually works.
Now suppose real yields climb and gold climbs too. The usual link has broken. That doesn’t mean gold is wrong. Maybe buyers are worried about something the bond market hasn’t priced yet. Maybe a central bank is buying. The gap doesn’t answer the question. It just tells you there is one worth asking.
Think of two dancers who’ve performed the same routine for years. You don’t need to know the choreography to notice when they fall out of step. You can’t tell who missed the beat — only that something changed. The wobble is the information.
To measure “out of step” you need a normal to compare against. That’s what a z-score does: it counts how many standard deviations today’s relationship sits from its own recent norm. Zero means business as usual. Beyond plus or minus two is unusual — roughly the outer edge of where the pair normally sits.
- Z-score — how far from normal, measured in standard deviations.
- Base rate — what actually happened after past episodes that looked like this one.
- Sample size (n) — how many past episodes there were. Small n, weak evidence.
- Edge — how much better (or worse) than the ordinary drift the outcome was.
Here’s the part most people skip. Knowing a pair is stretched is not the same as knowing what happens next. So the honest question is: after past episodes stretched this far, what did this asset actually do? That’s a base rate — a count of history, not a forecast. It can be reported. It cannot be promised.
And base rates deserve suspicion. When you test many relationships at once, some will look like they predict things purely by luck. Finance research has a long record of “edges” that vanished once people accounted for how many things had been tested to find them.[1]
There’s a second trap. Divergence episodes clump together in time — one stretched week is usually followed by another, because it’s the same episode still running. So forty “episodes” might really be five, seen forty times. The sample looks bigger than the evidence actually is.
None of this makes the signal useless. It makes it a clue rather than a verdict. A stretched pair with a clear economic story and a large sample is worth your attention. A stretched pair with a tiny sample and no reason behind it is probably noise wearing a costume.
Geopolitics is one reason several pairs can dislocate at once. Researchers at the Federal Reserve built an index that counts how often newspapers write about war, terrorism, and international tension, giving a measurable read on something that used to be pure vibes.[2]
Finisdom’s Macro Intel page tracks a dozen of these pairs — gold vs real yields, copper vs the 10-year, credit quality vs stocks, Treasuries vs stocks, the dollar vs emerging markets, and more. Each row shows its z-score against a roughly three-year norm, and opening a row shows the base rates at 5, 21, and 63 days with the sample size attached. It also tracks the four major central banks and the Caldara–Iacoviello geopolitical-risk index.