Multi-System Correlation Risk and Decorrelation

Running multiple systems does not diversify if those systems correlate. A portfolio of five trend-following variants is one bet, not five. Correlation risk is the hidden concentration that turns a "diversified" book into a single drawdown.

Measure Correlation Correctly

Compute pairwise correlation of daily returns across all systems over at least 200 trading days. Build the matrix. Two failures are common:

• Correlation of equity curves rather than returns. Equity curves always look correlated because they trend up. Use daily returns.
• Full-period average that hides regime correlation. Compute rolling 60-day correlation and watch the maximum, not the average. Systems that decorrelate in calm markets and correlate in crashes are the most dangerous.

Target: average pairwise correlation below 0.3. Maximum pairwise correlation in any stress window below 0.6.

Sources of Hidden Correlation

• Same edge, different instruments: two momentum systems on different symbols still correlate when market-wide trends reverse.
• Same risk trigger: systems that exit on the same volatility filter all stop out together.
• Common macro exposure: equity mean-reversion and equity trend systems both lose in a sharp gap down.

Audit each system's edge and risk logic. If two systems share an edge family or an exit trigger, expect them to correlate under stress even if their daily correlation looks low.

Decorrelation Techniques

1. Mix edge types: combine trend-following, mean-reversion, and carry. These edges have structurally different return profiles and lower stress correlation.
2. Mix timeframes: a daily system and a 5-minute system rarely draw down simultaneously.
3. Mix asset classes: equities, FX, commodities, and rates respond to different shocks.
4. Stagger entries: offset entry times so simultaneous signals do not cluster risk on one bar.

A target blend: no more than 40% of capital in any single edge family, no more than 50% in any single asset class, and at least three uncorrelated timeframes.

Allocation Rules

• Allocate by inverse volatility, not equal weight. A high-volatility system should not dominate portfolio risk.
• Cap single-system risk contribution at 25% of portfolio risk. If one system's volatility-weighted share exceeds 25%, cut its size.
• Rebalance monthly. Drift lets winners dominate and reconcentrates risk.

Stress Test the Portfolio

Run a Monte Carlo on the combined equity curve using the historical return correlation matrix, not independent shuffles. The 95th-percentile portfolio drawdown is your real risk number. If it exceeds 1.5x the sum you would expect from independent systems, correlation is biting.

The Failure Mode

The classic failure: a diversified-seeming book that draws down 25% in a single week because every system shared a hidden equity exposure. Correlation rises precisely in the weeks that matter. Design for the stress-window correlation, not the average, and the portfolio survives the episodes that break single-system traders.