Correlation Breakdown During Market Stress

Diversification tends to disappear in a broad selloff, because forced selling pushes unrelated assets to move together exactly when protection matters most.

Correlation breakdown during market stress means the diversification a portfolio relied on in calm periods disappears exactly when it is needed most, because a broad selloff pushes previously uncorrelated or negatively correlated assets to move together as investors sell whatever they can to raise cash. A risk model built on correlations estimated from a calm sample will systematically understate portfolio risk during exactly the kind of event that matters most, since the correlations it assumes are not the correlations that show up once the market is actually under stress.

The mechanism is largely about liquidity and forced selling rather than any change in the underlying fundamentals connecting different assets. When a fund faces margin calls or redemptions, it sells whatever it can convert to cash quickly, which is often its most liquid holdings rather than the positions most related to whatever triggered the initial shock. That forced selling pushes down assets with no fundamental connection to each other purely because they were held by the same distressed sellers, which is a distinct source of correlation from anything a covariance matrix estimated on fundamentals alone would predict.

The 2008 financial crisis saw correlations across equities, credit and even some historically defensive assets rise sharply as deleveraging spread across asset classes simultaneously. March 2020 produced a similar pattern over a much shorter window, with equities, corporate bonds and even government bonds selling off together for several days as investors raised cash indiscriminately, before central bank intervention reversed the move. February 2018 showed a narrower version of the same effect concentrated in volatility products, when a sudden spike in the VIX broke a set of short volatility strategies that had relied on a stable relationship between implied and realised volatility.

2022 is the example most relevant to traditional diversification specifically, since equities and government bonds fell together for a sustained stretch as rising rates hurt both asset classes at once, undermining the negative correlation that risk parity explained depends on to justify holding a levered bond allocation as an equity hedge.

Consider a two asset portfolio where the calm period correlation between equities and bonds is negative 0.3, but the stress period correlation rises to positive 0.5. Holding volatilities and weights fixed, portfolio variance under the stress correlation can run close to double what the calm period correlation would predict, since the diversification benefit that the negative correlation provided in the calm estimate is erased entirely and replaced with an added source of joint loss. A risk model that uses a single unconditional correlation estimated over several years will sit somewhere between these two figures and will underestimate the loss in exactly the periods that matter most for…