Correlation shifts: adjusting a multi-instrument book when the dollar strengthens
Treat portfolio risk as a set of correlations rather than a set of independent positions.
Why correlations change
Correlation between assets is not a constant; it depends on the driver currently dominating the market. When rate expectations lead, the dollar and gold typically move inversely. When geopolitical risk leads, the dollar and gold can strengthen together because they share a safe-haven character. Mid-October sits in the transition between those two regimes: CPI beating expectations lifted rate expectations, while Middle East supply risk lifted safe-haven demand.
The current correlation state
15 October was representative. After CPI, the dollar index spiked and gold lost 4,100 dollars — a classic negative correlation. But gold then rebounded quickly from 4,130 dollars while the dollar index gave back most of its gain into the close, weakening the strength of that negative relationship. Safe-haven demand is competing with the rates driver.
Three portfolio-level traps
First, positions that look diversified but are not. Being short EUR/USD and short GBP/USD at the same time is, in substance, one long dollar position and therefore one risk. Second, simultaneous stop-outs. When two instruments are highly correlated, a single event triggers both stops and the realised loss is the sum. Third, compounded margin usage. Highly correlated positions produce floating losses at the same time under stress, so margin requirements rise together.
How to adjust
Step one: compute rolling correlations between each position and the dollar index over the same window, say sixty trading days, and group anything above 0.7 into one bucket. Step two: set the total risk limit per bucket rather than per position. Step three: before an important release or event, check the combined floating loss of the portfolio under a stress scenario.
When to actively reduce correlation
When correlations rise to historic highs, diversification works least well and you should actively cut the number of positions in the same bucket. Conversely, when correlations are low and drivers are independent, diversification works best and you can hold more positions inside the same total risk limit.