Conditional correlation is the co-movement between assets measured within a specific market state or regime rather than over the full sample. It captures the empirical fact that correlations are not stable: cross-asset and cross-stock linkages typically rise sharply in stress states, so the relevant number is the one that holds during the drawdown, not the calm-period average.
Formally, it is the correlation of returns conditioned on an event or regime variable — e.g. correlation given the market is in its lower return tail. Unconditional (full-sample) correlation blends benign and stressed periods and understates joint downside risk; models such as DCC-GARCH estimate how the conditioning correlation evolves through time. The key asymmetry: correlations cluster toward one precisely when diversification is most needed.
With index ownership concentrated in passive vehicles and the Magnificent Seven dominating cap-weighted benchmarks, the conditional correlation of a future drawdown is higher than placid realised dispersion suggests — mechanical, flow-driven selling raises the odds that names fall together when the tape turns.
In the 2008 crisis, equity pairwise correlations that averaged roughly 0.3–0.4 in calm periods spiked above 0.8 as markets fell, and the same regime shift recurred in March 2020. A portfolio sized on the unconditional 0.35 would have been carrying far more joint downside than that single number implied once the stress state arrived.