Correlation Decay: When "Uncorrelated" Strategies Converge in Stress

A diversified portfolio of "uncorrelated" strategies looks fine in calm markets. The same portfolio's strategies often correlate to ~0.8 in crisis. This is not bad luck — it is structural, and any portfolio construction that ignores it under-prepares for the regimes that matter.

Why correlations rise in stress

Three mechanisms:

1. Common-factor exposure

Strategies that look uncorrelated in normal markets often share common-factor exposure that only manifests in stress. A trend-following book and a carry book might have correlation 0.1 over five years — but both are short volatility in a particular sense, and both lose money on the same vol-spike day.

2. Liquidity factor dominance

In normal markets, a strategy's return is driven by its idiosyncratic edge. In liquidity stress, the dominant factor becomes "is this position easy or hard to liquidate?" Every strategy holding hard-to-liquidate positions gets hit simultaneously. Apparently independent strategies converge.

3. Forced-sell mechanics

When a fund or prop desk reduces exposure due to drawdown limits, it sells everything proportionally — not just the losing strategies. This forces correlation across the entire book during the deleveraging window. If multiple desks hit drawdown limits simultaneously (which happens in crisis), correlation across the entire systematic universe spikes.

Empirical scale of the effect

Across the 2008 quant crisis, the average pairwise correlation between systematic equity strategies that historically traded near 0.0 jumped to ~0.7 within a single week. The effect persisted for months as deleveraging worked through the system.

The 2020 March COVID episode was similar but shorter — a few days of forced correlation followed by rapid recovery.

Implications for portfolio construction

1. Don't trust historical correlation matrices

A correlation matrix estimated on 5 years of normal markets understates stress-regime correlation by a factor of 3–5×. Risk-parity allocations computed on such a matrix systematically over-allocate to "diversifying" strategies that aren't.

2. Stress-test correlations explicitly

Re-compute correlations on the worst 5% of portfolio days. If those correlations differ materially from full-sample correlations, the portfolio has a hidden stress-regime concentration.

3. Diversify across stress signatures, not just across correlation in normal times

Two strategies might have correlation 0.0 in calm markets and correlation 0.8 in stress. Two other strategies might have correlation 0.3 in calm markets and correlation 0.3 in stress. The second pair is more diversified in any meaningful sense.

The relevant correlation is conditional: correlation given that drawdown is in the worst quintile.

4. Reserve liquidity for stress

In calm markets, the portfolio can run at full sizing. In stress, the portfolio's true risk concentration is much higher than nominal. Pre-commit to reduced sizing when stress regime indicators fire (see regime detection).

Where diversification still works

Across asset classes rather than across strategies within an asset class. Equity strategies vs rates strategies vs commodity strategies have more durable diversification than equity strategy A vs equity strategy B.
Across strategy archetypes with opposite regime sensitivities. Trend loves crisis; mean-reversion often hates it. Holding both is more durable than holding two trend variants.
Across capacity tiers. A strategy with $10M capacity has different liquidity dynamics than one with $1B capacity. Combining them buffers liquidity-stress correlation.

Practical takeaways

Correlation in calm is not correlation in crisis. Plan for the latter.
Conditional correlation (correlation in worst-decile drawdown) is the relevant input. Most allocators don't compute it.
Diversification across regime signatures beats diversification across statistical correlation. Always check both.