Correlation / Portfolio risk
Two positions in a USD 10,000 account each have USD 100 of planned stop-loss risk. Their combined budget is USD 200, or 2%; correlation does not change that sum. Separately, consider a model in which each position’s P&L has a USD 100 standard deviation over the same horizon. At correlation 0.8, the standard deviation of their sum is √(100² + 100² + 2 × 0.8 × 100 × 100) ≈ USD 189.74; at zero correlation it is about USD 141.42. This is modelled volatility, not a loss cap or a valuation of stops.
The direction of exposure matters
Long EUR/USD and long GBP/USD share exposure against USD despite different symbols. Shorting one pair reverses its contribution, but does not necessarily create an effective hedge: notionals, currencies and sensitivities can differ. For CFDs this is economic exposure, not a physical currency purchase. Mapping currency components and responses to a USD scenario can be more informative than merely counting the symbols held.
The measure depends on data and horizon
Describe historical correlation precisely: returns or position P&L, frequency, sample window and aligned timestamps. High correlation between the levels of two rising exchange rates is different from dependence between their changes. One sample does not guarantee a similar relationship in the next period. Ordinary correlation also fails to describe all joint extreme losses, price gaps and changing liquidity conditions.
Combine limits with scenario analysis
Before an entry, review planned losses, currency exposures, margin and shared event risks. Test adverse moves across all positions together with wider spreads or financing costs. Do not multiply stop-loss budgets by correlation to obtain “actual risk”. Low historical correlation does not justify unlimited increases in size. Portfolio risk needs several measures; no single table turns past relationships into a guarantee of future diversification.
Further learning and sources
9.9 Pair correlation, hidden exposure and portfolio risk; 7.8 Intermarket analysis in practice — combining signals.
