Currency Correlation and Hidden Risk
Why two 'different' pairs can be the same bet, the standard FX correlation patterns, the DXY factor that drives most of them, and what crisis-regime correlation breakdown does to retail accounts.
A retail trader running three trades at once feels diversified. Three pairs, three theses, three setups. The market structure underneath says something different. Most of the time, those three trades are the same bet measured three ways, and the trader has put three times as much capital behind one view as they think they have. When the view goes wrong, the loss is not three independent losses; it is one large loss in correlated clothing.
This is the single most common form of concealed risk in retail FX portfolios, and the cleanest way to think about it requires the same piece of arithmetic that quants and risk managers use: correlation. This article walks the FX correlation web from the ground up: what correlation is, the standard patterns among the majors, the DXY factor that explains most of them, why correlations break in crises, and how to size positions when you are running more than one at a time.
What correlation is, in one paragraph
Correlation measures how two series move together over a window. It runs from +1.0 (perfect lockstep) through 0 (no relationship) to −1.0 (perfect mirror). A correlation of +0.8 means the two series move in the same direction most of the time and by similar magnitudes; +0.3 means there is a tilt in the same direction but plenty of independent movement; 0 means knowing one tells you almost nothing about the other; −0.9 means the two are nearly mirror images.
Correlation is computed over a window (commonly 30 or 60 days), and the window matters. A rolling correlation that runs +0.6 over 60 days might spike to +0.95 in a crisis week. The static “long-run average” correlation hides a lot of regime-dependent behaviour.
The standard FX correlation table
The major-pair correlations are not random. They follow a pattern set by the underlying composition of each pair, and that pattern is extremely stable in calm conditions. A representative table of typical 60-day correlations against EUR/USD, in roughly normal market conditions:
| Pair | Typical 60-day correlation with EUR/USD |
|---|---|
| GBP/USD | +0.75 to +0.90 |
| AUD/USD | +0.55 to +0.80 |
| NZD/USD | +0.50 to +0.75 |
| USD/CHF | −0.85 to −0.95 |
| USD/CAD | −0.50 to −0.75 |
| USD/JPY | −0.10 to +0.40 (highly variable) |
| EUR/GBP | +0.10 to +0.40 (highly variable) |
| GBP/JPY | +0.30 to +0.65 |
| AUD/JPY | +0.40 to +0.70 |
| DXY | −0.85 to −0.95 |
The numbers shift with the regime, but the signs are stable: the USD-on-the-right pairs (EUR/USD, GBP/USD, AUD/USD, NZD/USD) are positively correlated with each other and negatively correlated with the USD-on-the-left pairs (USD/CHF, USD/CAD). The yen pairs are the outliers because the yen has its own driver (US real yields, carry flows) that is not the same as the broad dollar story.
The single factor explaining most of this
The unified explanation is the Dollar Index (DXY). Most of the day-to-day movement in the major pairs is explained by the dollar half of the quote moving, not the non-dollar half. A strong-dollar day pushes EUR/USD, GBP/USD, AUD/USD all down and USD/CHF, USD/CAD both up. The pairs look like seven different markets; they are mostly one market (the dollar) measured against seven different counterparties.
This is also why DXY is the single most-watched chart for any retail FX trader: it summarises the common factor that drives most of the correlation matrix. When DXY moves decisively, every USD pair moves with it. When DXY is flat, the correlation matrix loosens and non-USD stories show through.
The non-USD pairs (EUR/GBP, EUR/JPY, GBP/JPY, AUD/JPY, EUR/CHF) are expressions of the relative story between the two non-USD currencies, with the dollar factor washed out. They have their own dynamics, and their correlations with the USD pairs are less stable.
Three trades, one bet: a worked example
Consider a retail trader who opens, on the same morning, the following three positions, sized at 1% risk each:
- Long EUR/USD at 1.0850, stop at 1.0820.
- Long GBP/USD at 1.2700, stop at 1.2660.
- Short USD/CHF at 0.9000, stop at 0.9030.
Three pairs. Three pre-trade plans. Three different stops. The trader’s internal narrative is “I have three diversified trades, each risking 1%, total portfolio risk 3%”.
That is not what the math says. All three positions are short the dollar. EUR/USD long is structurally a short-dollar trade. GBP/USD long is the same. USD/CHF short is the same. The trader has put three times their per-trade risk behind one underlying view: “the dollar will weaken”. If the dollar strengthens for any reason (a hawkish Fed surprise, a flight-to-safety bid, a strong US data print), all three trades move against the position simultaneously, and the realised loss is close to 3% in one move rather than three independent 1% losses spread across uncorrelated events.
The size disaster is hidden because each pair has its own ticker and its own chart, and looks visually independent. The disaster is real because the correlation matrix says they are not. The trader’s effective per-bet risk was 3%, not 1%, and the risk framework was quietly broken before the first trade was opened.
This is the cleanest illustration of why naive trade-counting is the wrong unit. The unit is the underlying bet, and the bet is what the correlation matrix groups together.
Correlation isn’t constant
The numbers in the table are typical, not fixed. Correlations shift with the regime, and the shifts matter more than the long-run averages.
A short taxonomy of common regime shifts:
- Pure DXY regime. Macro story is dominated by US monetary policy or US-data surprises. Every USD pair moves to the same drumbeat. Correlations within the USD bloc are very high (positive between USD-right pairs, negative between USD-right and USD-left). This is the standard “the dollar is in charge” market.
- Risk-on / risk-off regime. Macro story is global risk appetite. EUR/USD, AUD/USD, NZD/USD, GBP/USD trade as risk proxies. USD/JPY and USD/CHF trade as safe havens. Cross-pair correlations strengthen along the risk axis and weaken along the pure-DXY axis.
- Local-story regime. A specific country has an idiosyncratic story (UK fiscal crisis, BOJ surprise, eurozone politics). The affected pair decouples from its usual partners. EUR/USD and GBP/USD can move opposite directions on the same day. Correlations drop sharply.
- Crisis regime. Liquidity dries up, carry trades unwind, flight-to-safety dominates. Most non-haven currencies fall against the dollar and the yen simultaneously. Cross-correlations spike toward extreme values. The portfolio that looked diversified in calm conditions reveals itself to be one bet.
The lesson is not “the table is wrong”. The lesson is “the table describes the average regime, and the average is not the regime you will be running risk in during a crisis”.
Crisis correlation: the most dangerous shift
A reliably observed pattern across asset classes: correlations rise in crises. The diversification that looks robust in calm conditions thins out in stress. In FX specifically, the crisis-week patterns are:
- All non-haven currencies fall against the dollar and yen. EUR, GBP, AUD, NZD, CAD, CHF (yes, CHF too in true panics), and most EM currencies move in the same direction at once.
- Carry-trade currencies sell off hardest. AUD, NZD, MXN, ZAR, TRY take the biggest hits as carry positions are forced to unwind.
- Cross-pair correlations approach +1. AUD/JPY and any other risk-sensitive pair move as one. A “diversified” book of three carry-style trades becomes a single trade with three times the intended risk.
- Bid-ask spreads widen. Liquidity providers protect themselves; the cost of exiting bad trades rises just when traders most need to exit.
This is the regime that turns a 3% portfolio-risk picture into a 12%+ realised loss in a week, when the correlation matrix the trader mentally relied on stops describing reality.
A practical sizing adjustment
The fix is not abstract. A small adjustment to position sizing puts the correlation issue inside the risk framework rather than next to it.
A workable rule of thumb:
If two or more open positions share the same underlying bet, size them so the combined risk equals your per-trade limit, not the sum of the per-trade limits.
In the worked example above, the trader’s three short-dollar trades should have been sized together at 1% total risk, not 1% each. That might mean a half-sized EUR/USD, a half-sized GBP/USD, no USD/CHF. Or it might mean picking the best of the three and skipping the other two. The decision is the trader’s; the constraint is the correlation arithmetic.
A more advanced version applies to the implied total exposure:
Compute the dollar beta of the open book (how much the portfolio moves for a 1% DXY move). If it exceeds your per-trade dollar risk at the typical DXY daily volatility, the book is over-concentrated.
Most retail platforms do not show this number, but it is computable from position sizes and pair-level correlations. Once you have done it a few times, the intuition takes over and the mental shortcut “how many dollar bets am I really running” becomes automatic.
What correlation isn’t
A few things correlation explicitly does not tell you:
- Causation. Two series can be correlated because they share a common driver, or because one drives the other, or by coincidence in a small sample. Correlation alone does not distinguish.
- The direction of the next move. Correlation describes co-movement, not prediction. Two pairs correlated +0.9 can both go up or both go down; correlation tells you they will likely agree on direction, not which direction.
- Robustness in crisis. The historical correlation is a useful prior, not a guarantee. The crisis-week behaviour discussed above can swamp the long-run average.
- The right sizing for an uncorrelated book. If your trades really are uncorrelated, the standard per-trade sizing is fine and the diversification benefit is real. The correlation adjustment only matters when the trades are not actually independent.
The takeaway
Two FX pairs that look like independent trades on a chart often share a single underlying bet, and that bet is usually some form of “the dollar”. The major-pair correlation matrix is stable in calm conditions, organised around DXY, and predictable enough that a short-EUR-long-CHF-USD setup is essentially three expressions of the same dollar view. Sizing those three at 1% each is sizing the underlying bet at 3%. Crisis regimes tighten correlations further; the diversification that felt real in calm conditions thins out exactly when it is most needed.
The fix is small and arithmetic: size the bet, not the trades. Group correlated positions, allocate the per-trade risk across the group rather than each member, and pay attention to which regime you are in. For the broader risk framework, the home article is Risk Management Basics. For the macro context that explains why most of the matrix exists in the first place, read The Dollar Index (DXY).