Two thirds of pairs show a statistically real response — but none of it is large
These relationships are real and they show up consistently — but they explain less than a tenth of what a currency does on a given day.
The strongest link in the entire study explains under 8 % of daily movement. The other 92 % is news, positioning, flows and noise.
How to use this
- Do not build a trade on these alone. Nothing here is a signal; it is context for a decision you are making for other reasons.
- The right use is expectation-setting: knowing that a VIX spike will hurt a long AUD/CHF position, or that gold carries extra risk into a Fed meeting.
- Be suspicious of anyone selling a system built on correlations this size, including us. If we had found a tradable edge here we would say so — we did not.
The detail
Across 29 instruments and four market series, 79 of the 116 relationships pass the conventional significance threshold. These are not artefacts.
But size matters as much as significance. The strongest relationship in the entire study is a correlation of about 0.28. Squared, that is under 8 % of daily variation explained. The rest is everything else: data releases, positioning, flows, liquidity, and noise.
Both things are true at once. Rates and volatility genuinely move currencies, consistently enough to show up clearly over five years. And on any individual day they will usually not be the reason your pair moved. A model built on these alone would be a poor model; ignoring them entirely would also be a mistake.
How many relationships are significant, by series
the numbers behind it RECOMPUTED 03 Sep 2026| Series | Pairs tested | Significant | Strongest correlation |
|---|---|---|---|
| US10Y | 29 | 21 | +0.240 |
| US2Y | 29 | 19 | +0.276 |
| US30Y | 29 | 20 | +0.153 |
| VIX | 29 | 20 | +0.210 |