Correlation measures how closely two markets move together. It runs from plus one, moving in lockstep, through zero, no relationship, to minus one, moving in exact opposition. It is calculated from returns over a chosen period and it changes over time, which is the first thing to know about it.
The common relationships
EUR/USD and GBP/USD usually move together, since both are the dollar against a European currency. EUR/USD and USD/CHF usually move in opposition. The Australian and New Zealand dollars move together and, loosely, with commodity prices. Gold tends to move against the dollar and against real interest rates. The US stock indices are strongly correlated with one another; the S&P 500, the Dow and the Nasdaq rarely disagree on direction over a day, though the Nasdaq moves further. European and Asian indices follow the US with a lag and a lower correlation.
Correlation is not fixed
These relationships hold on average and break when it matters. In a market panic, assets that normally move independently fall together, because everything is being sold. Correlations "go to one" in a crash, which is precisely the moment a trader hoped diversification would help. Any use of correlation has to allow for its worst case, not its average.
Why it multiplies risk
Two positions in markets that move together are, in effect, one larger position. Long EUR/USD and long GBP/USD, each risking one per cent, is closer to two per cent risked on one idea, the dollar falling, than to two independent one-per-cent bets. Long three US indices is one bet on US equities, three times over. Total open risk should be counted on that basis.
A trader has four positions open, each risking one per cent: long US500, long US30, long USTEC and long DE40. On the surface, four per cent at risk across four markets. In practice all four are long equities, and on a day when US stocks fall two per cent, all four positions move against the trader at once. Three of the stops are hit within the same hour. The trader has taken a single three-per-cent loss on one view, and the fourth position is close behind.
A multi-market EA should account for correlation in its sizing, either by treating correlated markets as one exposure or by limiting how many positions in the same family it holds at once. Its documentation should say which. An EA that treats a long in each of eight stock indices as eight independent trades is, in a bad week, taking one very large trade.
The other side is the useful one. Strategies whose results are uncorrelated with one another, trading different markets by different logic on different timeframes, smooth a combined record in a way that adding more correlated positions never can. That is the subject of the next page.