How to Size Correlated Trading Positions
Position risk cannot be assessed by examining each instrument in isolation when holdings respond to the same economic factors. Two individually modest trades can create concentrated exposure if their returns move together. Correlation and covariance provide a framework for studying combined variability, but historical estimates can change sharply during stressed markets.
Individual position size is not the whole picture
A trader might hold EUR/USD long and GBP/USD long, both involving US-dollar short exposure under common quote conventions. Although the currencies differ, a broad dollar rally may move both positions adversely. Similarly, several technology shares can share sensitivity to interest rates or semiconductor demand. The number of positions is not the same as the number of independent risks.
Portfolio variance equation
For two positions with portfolio weights w1 and w2, return volatilities sigma1 and sigma2, and correlation rho, variance equals w1²sigma1² + w2²sigma2² + 2w1w2sigma1sigma2rho. Weights and volatilities must use compatible currency and horizon conventions. The equation describes a statistical second moment, not maximum loss. Derivatives can create nonlinear exposure that ordinary return covariance inadequately captures.
Worked example: correlation increases risk
Imagine two equally weighted assets, each with annualised volatility of 20% and correlation 0.8. Portfolio volatility is the square root of 0.25×0.04 + 0.25×0.04 + 2×0.5×0.5×0.20×0.20×0.8, approximately 18.97%. If correlation were zero, volatility would be about 14.14%. The same weights have much less diversification benefit when the assets move closely together.
A currency exposure example
Suppose a portfolio is long £100,000-equivalent EUR/USD and £100,000-equivalent GBP/USD notional. The two positions involve different amounts of base currency but both can benefit from a weaker dollar. A stress scenario with EUR/USD down 3% and GBP/USD down 4% would create roughly £3,000 and £4,000 proportional gross exposure losses under a simplified common-currency approximation, before currency conversion and contract-specific rules. Exact P&L requires converting the positions and accounting for the quote conventions.
Correlation is not constant
In a market panic, risky assets that normally move separately may decline together. Historical covariance matrices estimated during calm conditions can understate stress losses. Asset relationships can also change when inflation shocks replace growth shocks as the dominant influence. Compare rolling correlations, event-based stress cases and common factor exposures rather than using one fixed coefficient.
Concentration by economic factor
A portfolio may have exposure to the dollar, oil prices, duration, equity beta, credit spreads and volatility simultaneously. Group positions by likely common drivers and consider how each responds to a selected scenario. Hedging one factor may introduce another: for example, a currency hedge can add funding or counterparty exposure. Model the actual trade direction and contract multiplier, not just a ticker’s label.
Sizing and volatility targets
An analyst can start from an aggregate portfolio risk budget and estimate how a proposed position changes expected volatility or scenario losses. Adding a position that is highly correlated with current holdings may provide little diversification. But volatility targets can encourage larger leverage during calm periods, exactly when a sudden regime shift may be most dangerous. Use leverage and liquidity limits independently of the covariance estimate.
A review checklist
List every instrument’s notional value, currency, direction and risk factor. Convert exposures to common reporting units, estimate correlations from appropriate data and stress multiple adverse scenarios. Consider gaps, nonlinear derivatives, margin calls and liquidity. Do not treat two separate stop-loss levels as proof that portfolio drawdown is capped.
For additional risk and market-structure context, refer to correlation and portfolio concentration.
Worked example: correlation changes combined risk
The following original example uses hypothetical inputs to illustrate the method; it is not an actual market quotation.
| Assumption | Value | Interpretation |
|---|---|---|
| Position A weight and annual volatility | 50%; 20% | Contribution to portfolio variance |
| Position B weight and annual volatility | 50%; 15% | Second component |
| Correlation +0.8 | Portfolio volatility ≈ 16.62% | Highly correlated returns limit diversification |
| Correlation 0.0 | Portfolio volatility = 12.50% | Uncorrelated fluctuations diversify |
| Correlation −0.5 | Portfolio volatility ≈ 9.01% | Partial offset between return fluctuations |
The simplified two-asset calculation is σp² = wA²σA² + wB²σB² + 2wAwBσAσBρ. For the highly correlated example, that is 0.25(0.20²) + 0.25(0.15²) + 0.5(0.20)(0.15)(0.8) = 0.027625; its square root is approximately 16.62%. Historical correlations are not constant, and realised portfolio volatility can differ materially from an estimate based on yesterday’s data. The exercise illustrates why separate tickets are not necessarily separate risks.
Frequently asked questions
Does holding two positions halve risk?
No. Highly correlated positions may provide little diversification.
Can correlation change during a crash?
Yes. Stress conditions can create stronger co-movement.
Does low historical volatility imply safe leverage?
No. Large jumps and regime changes can exceed historical estimates.