Correlation exposure

Definition

Correlation exposure is the risk that positions sized as independent turn out to be one position: when correlations rise, a portfolio of many small risks becomes a single large risk. Its central empirical fact is asymmetry — correlations between risk assets increase in falling markets, precisely when diversification is being relied upon. Sizing that ignores this is systematically overconfident.

How it works / structure

• Measurement (engine-executable): rolling pairwise correlations per position pair (window pinned); portfolio effective-position count (equity spread across n assets at average correlation ρ behaves like roughly n / (1 + ρ(n−1)) independent positions — the platform’s aggregation statistic); factor decomposition (positions regressed on common factors — strategy-factor-investing machinery — to expose shared drivers that pairwise correlation understates).
• Stress convention: risk aggregation uses STRESSED correlations (elevated toward historical crisis levels), not calm-period estimates — the Longin-Solnik/Ang-Chen asymmetry built into the arithmetic (risk-scenario-analysis runs the scenarios).
• Budget interface: port-correlation-budgets sets the allowable aggregate; this entry is the measurement layer.

When it applies

Any book with more than one position — which is every book. The canonical applications: many-small-positions strategies whose per-trade risk claims assume independence (risk-fixed-fractional’s aggregation caveat), hedged structures whose hedge is a correlation assumption (strategy-pairs-trading unwinds), and cross-asset books whose “diversification” is calm-period correlation.

Risk profile & failure modes

• Crisis convergence: the documented failure — equity sectors, credit, international markets, and carry-shaped strategies converge toward high correlation in stress; portfolios built on calm-period matrices discover their true position count in drawdowns.
• Estimation instability: correlation estimates from short windows are noisy, from long windows stale; regime shifts (regime-volatility) move true correlations faster than estimators track.
• Factor blindness: twenty uncorrelated-looking stock positions sharing a rate-sensitivity or crowding factor are one trade wearing twenty tickers; pairwise matrices miss what factor decomposition catches.
• Netting illusions: long/short books net to small delta but can carry large correlated spread risk (port-exposure-netting).

Evidence & limits

Longin-Solnik (2001) showed international equity correlations rise significantly in bear-market tails (rejecting the constant-correlation model in the loss tail specifically); Ang-Chen (2002) documented the same downside asymmetry across US equity portfolios. This asymmetry is among the most consequential documented facts in portfolio construction. Precise stressed-correlation values are estimation choices, pinned per replay; the direction of the adjustment is evidence-forced.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Portfolio P’s effective position count, computed with stressed correlations, stays above 5 this quarter” — falsified by the computed series.
• “In the next 10%+ index drawdown, P’s realized average pairwise correlation will exceed its calm-period average by at least 0.2” — falsified by the measured comparison.

Cross-references

• Budget layer: port-correlation-budgets; the math it disciplines: port-diversification-math
• Scenario machinery: risk-scenario-analysis
• Illusions it audits: port-exposure-netting, strategy-pairs-trading, risk-fixed-fractional (aggregation)
• Regime driver: regime-volatility