Currency Correlation
Also known as: pair correlation, FX correlation, correlated pairs, positive and negative correlation
What is it?
Currency correlation measures how closely two pairs move together, on a scale from +1, where they move in lockstep, through 0, where there is no relationship, to -1, where they move exactly opposite. It matters because correlated positions stack risk without looking like it. EUR/USD and GBP/USD typically run a 30-day correlation near +0.85.
30-day correlation: where three separate trades are really one bet
5×5 matrixEUR/USD and GBP/USD at +0.86 are close to the same trade. USD/CHF at -0.90 is that trade inverted. Long both euro pairs and short USD/CHF is one dollar position, not three.
If you risk 1 percent on a long in each, you have not taken two independent 1 percent bets on two ideas. You have taken close to one 2 percent bet on the dollar falling, and when the dollar rallies both stops are hit by the same move. Negative correlation cuts the other way.
EUR/USD and USD/CHF sit near -0.90, so a long in each largely cancels out and you pay two spreads to hold something close to no position. Correlations are also not fixed. They drift with the rate cycle, and in a risk-off shock pairs that normally look unrelated converge toward +1 as everything trades against the dollar at once, which is exactly when the stacking hurts most.
Why it matters: Correlation is what quietly turns several small, independent-looking positions into one large concentrated bet that your per-trade risk limit never sees.
Correlation = covariance(A, B) / (standard deviation of A x standard deviation of B), giving a value from -1 to +1
Correlated positions multiply real exposure while every individual trade still looks correctly sized, which is how accounts breach a risk limit without a single oversized trade.
Real-world example
A trader long EUR/USD, long GBP/USD and short USD/CHF at 1 percent risk each believed they held 3 percent of risk across three ideas. All three were dollar-short at roughly +0.85, so one dollar rally took close to the full 3 percent in a single move.
How SignalBots handles it
SignalBots publishes signals across many pairs, so before taking two at once you can check whether they are the same directional bet on one currency and size them as one position. See /risk-warning.
Pro tip
Cap total exposure per currency, not per pair. Three open trades that are all short the dollar should count as one dollar position against that cap.
Common pitfalls
Reading a correlation once and treating it as permanent, when the figure shifts with the rate cycle and collapses toward +1 during risk-off shocks.
Frequently asked questions
What correlation counts as high?
As a working rule, above +0.70 or below -0.70 means the two pairs are effectively expressing the same view. Between -0.30 and +0.30 they are close enough to independent to size separately. The band in between deserves a reduced combined position.
What lookback period should I measure over?
Match it to your holding period. A day trader cares about the 20 to 30 day correlation; a swing trader holding for weeks should also check the 90 day figure, because a short window can show a relationship the longer record does not support.
Does negative correlation hedge my risk?
Partly, but it also cancels your edge. Holding EUR/USD and USD/CHF long at -0.90 mostly nets to flat while you still pay both spreads and both swap charges. A hedge is only useful when it is deliberate and sized on purpose.
Why do correlations break down?
Because they are a by-product of what is currently driving price, not a rule. When one central bank diverges or a local shock hits one economy, that currency stops following the shared driver and the relationship weakens or flips.
How does this apply to an automated strategy?
Run the same bot on EUR/USD and GBP/USD and you have not diversified, you have doubled one position. Cap the number of simultaneously open trades that share a currency, and backtest the strategy with that cap in place. Your capital is at risk.
Trading involves substantial risk of loss. Historical and backtested results do not guarantee future performance. Read the full risk warning.
