Open long EUR/USD, long GBP/USD and short USD/CHF, and you might feel diversified — three pairs, three ideas. You actually have one idea, three times: short the dollar. Correlation is the invisible wiring between pairs, and not knowing it means not knowing your own position size.
Where correlations come from
Shared currencies, by arithmetic. Pairs sharing a leg are mechanically linked. EUR/USD and GBP/USD share the dollar side, so any dollar-wide move pushes them together — historically these two spend most of their time strongly positively correlated. EUR/USD and USD/CHF share the dollar on opposite sides, making them strongly negatively correlated. And crosses are literally built from majors: EUR/GBP is EUR/USD divided by GBP/USD, so those three can never move independently — one of them is always implied by the other two.
Shared stories, by economics. AUD and NZD track each other because both are commodity-linked, Asia-exposed economies. CAD tracks oil. In risk-off episodes, the yen and franc strengthen together while high-beta currencies fall together — correlation across the whole board tightens exactly when you least want it to, because one factor (fear) is driving everything.
The risk-management consequence
Correlated positions add up, not average out. Three 1%-risk trades that are 0.9-correlated behave like one trade risking nearly 3% — you have not diversified, you have tripled. The discipline that follows: count risk per theme, not per ticket. If your book is long EUR/USD, long GBP/USD and long AUD/USD, write it in your journal as "short USD, ~3% at risk" and ask whether you would knowingly put 3% on a single dollar view. Sometimes the answer is yes — then at least it is a decision rather than an accident.
The mirror trap: hedging that isn't. Long EUR/USD and short GBP/USD feels like offsetting dollar exposure, but what remains is a pure EUR/GBP position — you have not reduced risk, you have changed the trade. Any time two of your positions "protect" each other, work out which cross you are actually holding.
Reading correlation numbers honestly
Correlation coefficients run from +1 (lockstep) through 0 (unrelated) to −1 (mirror image), computed over some lookback window — and the window is everything. A 30-day correlation answers "lately"; a 1-year answers "usually"; and the two frequently disagree. Three honesty rules: correlations are not stable — the AUD/NZD relationship that held for six months can break on one central-bank surprise, and every correlation table is a photograph of the past; tight windows whipsaw — short-lookback numbers swing wildly and invite overreaction; and correlation is not causation or lead-lag — two pairs moving together does not tell you which (if either) moves first, so "EUR/USD moved, so GBP/USD must catch up" is a coin flip dressed as analysis.
Putting it to work
Three practical uses survive the caveats. Exposure accounting — the theme-counting above, which is the big one. Confirmation — when your EUR/USD short thesis is a dollar thesis, a glance at whether USD/JPY and DXY agree tells you if the dollar move is real or a euro-only wobble. Instrument selection — if two pairs express the same idea, take the one with the cleaner chart and cheaper spread instead of both, or split the size across them if you must, never doubling it.
The simplest possible summary: before adding any position, ask what it does to your existing book if the dollar — or risk appetite — moves 1% overnight. If the honest answer is "everything I own moves the same way," you have one trade. Size it like one trade.