A multi-asset portfolio carries 12 positions across equities, bonds, and commodities. The covariance matrix shows average pairwise correlation of 0.22 — diversification looks textbook-perfect. Then March 2020 arrives. Within nine trading days, that average correlation jumps to 0.81. The portfolio loses 34% peak-to-trough. The maths were correct. The assumption about the maths was catastrophically wrong.
This is correlation breakdown — the documented tendency for asset correlations to converge toward 1.0 precisely when diversification is needed most. RiskMetrics research and BIS Working Papers both identify it as a primary reason why Value-at-Risk models systematically underestimate tail losses. A covariance matrix estimated from calm-period data is a liability masquerading as risk management infrastructure.
The mechanics are well-documented in academic finance literature. During stress, liquidity evaporates simultaneously across asset classes. Forced selling by leveraged participants becomes the dominant price driver — not fundamentals, not individual asset dynamics. When one factor (margin calls, redemptions, fear) drives everything, correlations mechanically compress toward unity. A 0.25 calm-period correlation between ASX equities and investment-grade credit becomes 0.79 under stress.
Practitioners address this using regime-switching covariance models. The approach involves estimating two separate covariance matrices — one calibrated to low-volatility periods, one to high-volatility or crisis periods — then weighting them using a Hidden Markov Model or simple VIX-threshold trigger. Position sizing is then constrained by the stress-period matrix, not the calm one. If a 2% portfolio risk budget is allocated per position under calm conditions, the stress matrix might cut that to 0.8% — without changing conviction in the trade. For deeper grounding in the statistical architecture, covariance and its portfolio implications are well-documented, the formal structure of statistical correlation and its limitations explains why the measure breaks down, and BIS research on Value-at-Risk model failures during crises provides the institutional evidence base.
The pre-crisis covariance matrix isn't wrong — it's answering a question nobody asked during a crash. Build for the regime you fear, not the one you're comfortable in.
This content is for educational purposes only and does not constitute financial product advice. Past performance is not indicative of future results. Profit Logic Ltd (ACN 688 669 936) accepts no responsibility for errors or omissions in this content or anywhere on this website. Always seek advice from a licensed financial adviser before making investment decisions.