ARMFPair correlation and the total risk of positions
Pair correlation and total risk are directly linked: two positions at 1 % are not always 2 % and almost never 1.4 %. The value depends on how linked the instruments are, and on the currency market the link is built into the very construction of the pairs. We take apart how to measure correlation and how to add risks correctly.
Where the link between pairs comes from
A currency pair is a ratio of two currencies, so any two pairs with a shared currency are linked mechanically. EUR/USD and GBP/USD share the dollar in the quote: when the dollar strengthens, both go down. EUR/USD and USD/CHF hold the dollar on opposite sides, so they mostly move in opposite directions.
- A shared currency on the same side of the quote
- The pairs behave alike. EUR/USD and GBP/USD, AUD/USD and NZD/USD — the moves often coincide in direction.
- A shared currency on different sides
- The pairs behave in opposite ways. EUR/USD and USD/CHF — a rise in one is usually accompanied by a fall in the other.
- No shared currency
- The link is not mechanical but can arise through common factors: risk appetite, commodity prices, central bank rates.
- The link is not constant
- Correlation changes over time and especially sharply in moments of strong moves. It has to be calculated on your own data and updated rather than taken once from a ready table.
How to calculate currency pair correlation yourself instead of using a ready table
Ready correlation tables go out of date, and the calculation period in them usually does not match your horizon. Your own calculation takes a few minutes in any spreadsheet.
For both pairs over the same period: 60–100 values on the timeframe you trade.
step 1Not the prices themselves but the changes: (current − previous) ÷ previous. The correlation of price levels is misleading — it is almost always high.
step 2In any spreadsheet this is one formula over the two columns of increments. The result is a number from −1 to +1.
step 3Once a month and after major events. A value calculated half a year ago describes a different market.
step 4Interpretation: values above +0.7 mean positions in the same direction on these pairs almost duplicate each other; from −0.3 to +0.3 the link is weak; below −0.7 positions in the same direction cancel each other out, while positions in opposite directions become one bet again.
The total risk of a portfolio of positions: how to add it up
Two different values are needed here, and they are often confused.
The spread formula for two positions: √(r₁² + r₂² + 2 × ρ × r₁ × r₂), where ρ is the correlation coefficient. For two positions of 1 % it gives the following.
| Correlation | Spread of the result | The maximum loss | What it means |
|---|---|---|---|
| +1.0 | 2.00 % | 2.00 % | One position of double size |
| +0.9 | 1.95 % | 2.00 % | Practically a duplicate: the typical link of pairs sharing the dollar |
| +0.5 | 1.73 % | 2.00 % | Partial duplication |
| 0.0 | 1.41 % | 2.00 % | Independent positions |
| −0.5 | 1.00 % | 2.00 % | The positions partly cancel each other out |
The key difference. Correlation reduces the typical swing of the account but does not reduce the maximum loss. Stops can trigger simultaneously at any correlation — with a weak link it simply happens less often. That is why the total risk limit is counted for the worst case, that is by plain addition.
The rule for dividing risk between linked pairs
The correlation calculation is needed for one decision: how to divide the risk when pairs are linked. The practical approach is to treat a group of linked positions as one trade and divide the ordinary risk between them.
| Correlation | How to count the positions | Risk on each at a 1 % limit |
|---|---|---|
| above +0.7 | As one trade | 1 % ÷ the number of positions |
| from +0.3 to +0.7 | As one and a half trades | about 0.7 % on each |
| from −0.3 to +0.3 | As independent | 1 % on each |
| below −0.7 | The positions cancel each other out | the risk is counted on the difference rather than the sum |
The last row is about a construction like buying EUR/USD and buying USD/CHF at the same time. Formally these are two trades, economically an almost neutral position in which you pay a double spread for the right to watch the difference in behaviour between two pairs.
A simple check with no calculations. Write out the currencies of all open positions and count how many times each one appears and on which side. A currency that repeats on the same side in three positions is your real bet, and the risk on it adds up.
Frequently asked questions
Which currency pairs are the most correlated?
All pairs with a shared currency are linked mechanically: EUR/USD and GBP/USD, AUD/USD and NZD/USD move alike, EUR/USD and USD/CHF in opposite ways. The specific values depend on the period and change, so they are calculated on your own data rather than taken from a static table.
How should correlation be taken into account when calculating size?
A simple working technique: if the correlation is above 0.7, treat the positions as one trade and divide the risk between them. Two positions at 0.5 % instead of two at 1 % give the same total risk as one full trade.
Does correlation change during crises?
It does, and usually in the inconvenient direction: in moments of strong moves the link between instruments grows. Sets of positions that looked independent in a calm market start moving in sync exactly when it costs the most.
Over which period should currency pair correlation be calculated?
Over the one you trade: for intraday trading, on hourly increments over the last few weeks; for swing trading, on daily ones over two or three months. A value from another horizon describes another market.
Why can a ready correlation table not be used?
Because it states neither the period nor the timeframe, while the link between pairs changes. Your own calculation takes a few minutes in a spreadsheet and gives a number that belongs to your horizon.
Should correlation be calculated on prices or on increments?
On increments. The correlation of price levels is almost always high and says nothing about joint movement: two rising series will show a link even if their swings are independent.
What to do with pairs whose correlation is near zero?
Treat the positions as independent and apply the full risk per trade to each within the overall limit. But check periodically: zero correlation is a property of the period, not of the pair.
Can negative correlation be used to reduce risk?
Deliberately, yes: opposite positions on linked pairs reduce the total bet on a currency. But that is already a separate construction, and its risk is counted from the difference in behaviour of the pairs rather than as the sum of two trades.
How quickly does a correlation calculation go out of date?
Noticeably within a month or two, and in periods of strong moves within days. A practical regime: recalculate once a month and always after events that change the behaviour of a currency — central bank decisions or sharp dollar moves.
Does correlation affect the choice of size?
It does, through the division of risk. At a correlation above 0.7 a group of positions counts as one trade: the ordinary risk is divided between them rather than applied to each.
Read next
