Diversification arguments usually rest on a single number: the average correlation between two assets over some historical window. If it’s low, the assets are described as diversifying.
The trouble is that averages describe ordinary conditions, and diversification matters most in extraordinary ones. An asset that moves independently in calm markets and moves with everything else during a sell-off has provided diversification precisely when it wasn’t needed.
Testing the case properly means looking at correlation in the tails rather than in the middle, and the picture there is different.
The Portfolio Question Behind the Definition
Beyond whats cryptocurrency as a definition, the portfolio question is narrower: does adding it change the risk profile of everything else?
That question has several components:
- Long-run correlation with existing holdings, measured over years
- Rolling correlation, which shows whether the relationship is stable or drifting
- Downside correlation, measured only across periods when both assets fell
- Correlation during specific stress events, rather than across an averaged sample
- Whether the relationship has changed as the holder base has changed
The last point turns out to matter more than most analyses allow for.
What Happens to Equity Correlation Over Time
The relationship with equity indices hasn’t been stable, and the direction of travel is documented.
Research applying rolling window correlation and event study methods to daily data from 2018 to 2025 found a significant increase in correlation following institutional milestones such as the introduction of exchange-traded funds, with correlations against major US indices peaking at 0.87 in 2024, and trends fluctuating across market regimes.
There’s a logic to that. As an asset moves into regulated fund vehicles held by allocators who also hold equities, it becomes subject to the same portfolio-level risk decisions. When those allocators reduce risk, they reduce it across everything they hold.
Institutional adoption is usually presented as a maturing influence. On this measure it has also made the asset behave more like the things it was supposed to diversify against.
Downside Correlation Versus Upside
The sharpest finding concerns asymmetry, and it comes from within the asset class rather than between it and equities.
Research examining tail dynamics over a decade found a persistent and widening downside asymmetry, with lower-tail exceedance correlation at 0.847 against upper-tail correlation of 0.246 at the 90th percentile, and late-period upper-tail correlation turning negative at −0.175 at the 95th percentile, implying that diversification within the cryptocurrency asset class remains illusory during market stress.
Read that carefully, because it’s a specific claim. The two largest assets in the category move together strongly when both are falling and much less strongly when both are rising.
The same research found genuine evidence of maturation elsewhere, with tail risk measures declining over the period. But the improvement concentrated in low-uncertainty periods, and drawdown behaviour in high-uncertainty regimes showed no significant change.
What a Real Diversifier Requires
The bar is higher than low average correlation:
- Low correlation during drawdowns, not just on average
- A different underlying driver, so the two assets respond to different variables
- Sufficient liquidity under stress, since a diversifier that can’t be sold isn’t one
- Stability of the relationship, or at least a known reason for it to change
- Enough size in the portfolio to matter, which conflicts with volatility limits on sizing
Very few assets clear all five, which is worth remembering before judging any single candidate harshly.
How to Test the Case Yourself
The calculation is accessible with public data and a spreadsheet:
- Pull daily returns for the holding and for the rest of the portfolio
- Calculate correlation across the full sample, as a baseline
- Recalculate using only days when the portfolio fell, which gives downside correlation
- Compare the two figures, since the gap is the part that matters
- Repeat over rolling windows to see whether the relationship is drifting
Doing this on a specific holding produces a more relevant answer than any general finding, because it uses the actual portfolio rather than a representative one.
The result also tends to be more stable than expected in one direction and less in another. Full-sample correlation moves slowly. Downside correlation can shift substantially within a single stress episode, which is why a figure calculated two years ago may describe a relationship that no longer holds.
What the Evidence Supports
None of this argues the asset class has no place in a portfolio. It argues that the diversification case specifically is weaker than the average-correlation figure suggests, and that anyone holding it for that reason should check the number they’re relying on.
Other reasons to hold it remain untouched by this evidence. Exposure to a growth thesis, a view on adoption, or a long-horizon position all stand or fall on their own merits.
What the research does establish is that the position is unlikely to cushion a portfolio during the weeks when cushioning would matter most, and sizing it as though it would is the specific error worth avoiding.
