When Everything Goes To One

A portfolio's risk model usually assumes that different positions don't all move for the same reason at the same time — that's the entire basis for holding many uncorrelated bets instead of one big one. In a genuine crisis, that assumption fails all at once. Stocks across every sector, currencies, credit, even assets that had nothing to do with each other historically, start falling together, because everyone is selling for the same reason: to raise cash, to cut risk, to meet a margin call. Traders call this "everything going to one" — pairwise correlations across the book rushing toward 1.0 right when a portfolio most needs them to stay low.

Why it happens

Correlation breakdown in a crisis isn't really about the assets themselves changing their relationship to each other; it's about a single common force — forced deleveraging — temporarily overwhelming every asset-specific reason they'd normally move differently. When funds across the market are forced to reduce risk simultaneously, they sell whatever is liquid enough to sell, not necessarily what they'd choose to sell in calm conditions. That means the assets that get hit hardest are often the ones that are easiest to trade, regardless of their fundamentals, which is precisely why unrelated positions start moving together: they're all being sold for the same liquidity reason, not for reasons specific to each one.

What this means in practice

A book that was sized assuming, say, an average pairwise correlation of 0.2 between its positions can behave, in a crisis, like a book with correlation of 0.7 or 0.8 — meaning the actual portfolio volatility realized is far higher than the risk model predicted going in, sometimes by several multiples. This is why stress tests and scenario analysis matter more than the standard covariance-based risk model during exactly the periods that matter most: the historical correlation matrix a risk model is built on is measured mostly from calm periods, so it systematically understates how correlated the book can get. A common practical response is to run the portfolio through an explicit high-correlation scenario — what does P&L look like if every position's correlation to every other jumps to, say, 0.8 — rather than trusting the standard estimate to capture that risk on its own.

The good news is that correlation spikes toward one are usually temporary; as forced selling exhausts itself, relationships tend to loosen back toward their historical norms. But "usually temporary" is not the same as "safe to ride out" — a book that's overleveraged for a correlation-of-one world can be forced to liquidate before the reversion happens, turning a temporary statistical anomaly into a permanent loss.