Correlation and portfolio risk

Two bots can look diversified by name while taking the same macro risk. Learn how to measure shared exposure before combining systems.

Different symbols can share one driver

US indices, sector instruments, currencies, and commodities can respond to common risk sentiment, rates, liquidity, or news. Holding several positions does not automatically spread risk if they tend to lose together during the same environment.

Correlation is not constant. A pair that appears independent in calm data can become strongly linked during a shock.

Measure portfolio outcomes, not labels

Combine the actual trade or daily return series from the systems you plan to run. Review correlation, worst combined days, overlapping exposure, simultaneous losing streaks, and contribution to drawdown. Use both normal and stressed windows.

Track gross and net exposure by market theme.
Measure maximum combined loss during a single session.
Cap the number of correlated positions allowed at once.
Test what happens when all systems experience their bad regime together.

Allocate risk deliberately

Portfolio allocation can be equal risk, volatility scaled, conviction weighted, or constrained by account rules. The method matters less than having a written rule that stops one correlated cluster from consuming the entire loss budget.

If two systems use the same data, session, and direction, treat that shared dependency as a portfolio risk even when the codebases are different.

Discuss this article

Ask a setup question, share a backtest, or compare notes with other algorithmic traders. Use a display name; your email remains private.

Create an account or sign in above to join the conversation.

Loading discussion…