# When Correlation Breaks: Diversification in a Drawdown

URL: https://www.caijing188.com/correlation-breaks/
Published: 2026-10-27T20:56:43+08:00 | Updated: 2026-10-27T20:56:43+08:00
Verified: 2026-10-27

## Quick answer
Correlation falls in calm markets and rises in a sell-off. Size on the drawdown figure, not on the full-sample average.

## Summary
Correlation is a measurement over a window, not a stable property. In a drawdown majors converge, and portfolios sized on calm-market data discover the gap too late.

## FAQ
Q: Does a low crypto correlation mean diversification works?
A: It means the two series did not move together in the sample you measured. It does not mean they will not move together in the next drawdown. Check the coefficient on falling days before concluding anything.
Q: Why do major assets converge so sharply?
A: Because the marginal seller is shared. Liquidations, redemption selling and collateral stress hit the most liquid assets first, and those are usually the majors. The convergence is a funding event, not a change in fundamentals.
Q: How often should correlation be recalculated?
A: On a fixed schedule, monthly at minimum, and immediately after any move big enough to trigger liquidations. A matrix older than the last stress event is a historical document.
Q: Can stablecoins reduce correlation risk?
A: They remove price exposure, which is a different thing from removing risk. They introduce issuer, reserve and redemption risk instead. Usually the right trade for a reserve, but a substitution of risks rather than an elimination.
Q: Is a negative correlation useful?
A: Only if it is stable and the instrument genuinely offsets the exposure. Negative coefficients measured over short windows flip often, and trading a flimsy one costs more in fees and slippage than it saves in variance.
Q: What is the simplest practical defence?
A: Assume a coefficient of one when sizing, then check whether the book survives. If it does not, the portfolio depends on a statistical relationship never tested in the conditions that matter.