Portfolios are often built on the assumption that different assets move independently. That assumption weakens in exactly the conditions where diversification is supposed to matter most.
Correlation measures the past, not a property
A correlation figure describes how two assets happened to move together over a chosen period. It is a summary of history rather than a fixed characteristic of either asset.
Change the period and the figure changes with it. A pair that looks unrelated over a decade of calm can look closely linked across a few violent weeks.
Because the number is backward looking, a portfolio built on it is calibrated to conditions that have already happened rather than to the ones ahead.
Forced selling overrides fundamentals
When leveraged positions come under pressure, the seller does not choose what to sell on merit. The seller sells what can be sold quickly at an acceptable price.
Liquid, profitable holdings are often the first to go, precisely because they can be turned into cash without moving the market far.
The effect is that unrelated assets fall together, not because their prospects have changed but because the same holders are liquidating them for the same reason.
Metals are not exempt from the mechanism
Gold has historically held up better than most assets in prolonged crises, but it can still fall sharply during the acute phase of one.
A holder facing a margin call and holding gold has an asset that is easy to sell at a fair price, which makes it a natural source of cash.
The pattern that has often followed is a fast decline alongside everything else, then a recovery as forced selling exhausts itself and buyers return.
Time horizon changes what the data says
Measured daily, the relationship between metals and equities during a crash can look uncomfortably close.
Measured over quarters or years, the same episode frequently shows the metal recovering while other assets are still repairing losses.
Which picture is relevant depends on whether a holder can wait, and that is a question about circumstances rather than about the asset.
Diversification is about causes, not statistics
Assets that share a common driver will eventually move together whatever the historical correlation suggests, because the driver reasserts itself under stress.
Holdings that respond to genuinely different forces stand a better chance of behaving differently, though no combination removes the effect of a general scramble for cash.
Understanding why an asset would rise or fall gives a more durable expectation than a coefficient calculated from a period that has ended.