In calm markets your assets behave the way their history says they should. Crypto drifts somewhat independently of equities. Gold drifts somewhat independently of both. Prediction markets on non-financial events are genuinely off doing their own thing. Then a crisis hits and everything you own is red at the same time, and the diversification you thought you had turns out to be a fair-weather friend.
Why correlations spike in crises
The mechanism is boring and it's always the same: forced selling. A leveraged fund gets a margin call and sells whatever it can, not whatever it wants to. Retail panics and dumps the whole account. Liquidity thins out and market makers pull quotes across every asset at once. For a short, intense stretch the thing setting prices isn't fundamentals or technicals, it's the plumbing of people being forced out of positions.
That's when assets with no fundamental connection can see their correlation jump to 0.8 or higher, purely because the same sellers are hitting the bid on all of them in the same hour. March 2020 is the clean example. Bitcoin, equities, gold, even Treasuries all dropped together in the first panic. They only separated once the acute selling burned itself out, which took about two weeks. For those two weeks diversification did nothing for you.
Measuring tail correlation
Plain Pearson correlation is dominated by normal days, so it tells you almost nothing about the tails. The number that actually matters for risk is tail correlation, the correlation during the worst 5 to 10 percent of market days. That's the only regime where you find out whether your portfolio is really diversified.
You can get at it by filtering your return series down to days where at least one asset dropped more than 2 standard deviations, then computing correlation on just those observations. It comes out higher than the full-sample number almost every time, sometimes a lot higher. Build your portfolio around that figure instead of the average and you get an honest read on how diversified you are when it counts.
What still diversifies under stress
A few things hold up when everything else is correlating.
- Long-dated Treasuries rally into equity selloffs in most environments, though 2022 was a loud reminder that "most" isn't "always".
- Managed-futures strategies that can go short tend to do well in extended drawdowns because they ride the trend down instead of fighting it.
- Cash is the ultimate crisis diversifier. Zero correlation with everything by definition, and its value survives whatever happens to risk assets. The return you give up holding it in a bull market is just the premium you're paying for protection, so it's worth treating cash as insurance rather than laziness.
- Prediction-market positions in non-financial contracts stay uncorrelated because the underlying events genuinely don't care about markets. A contract on whether it rains in London next Tuesday has never once checked the S&P 500.
Decide before the panic
The time to plan for a correlation spike is while you're calm and thinking straight, which is never during the spike itself. Write down specific actions tied to specific levels of stress. Something like: down 10 percent, cut every leveraged position in half. Down 20 percent, close the leveraged book entirely and sit in the core portfolio.
Then commit to running those rules mechanically, because in the moment your emotions will argue with every one of them. Having the decisions made in advance takes the choice off your plate at the exact time you're worst at choosing. I keep mine written down for the same reason a pilot uses a checklist.
And when it passes, correlations drift back to normal, which is its own opening. Names that fell together despite having nothing in common re-diverge on the way out. If you buy the ones with the strongest fundamental recovery case while correlation is still pinned high, you pick up the bounce and the diversification benefit as things renormalize. Worth watching for on the other side of the next one.