The question of what crypto moves with has two confident answers that contradict each other, and both are supported by real data from real periods. That is not a sign that one side is wrong. It is a sign that the relationship is not stable, and knowing that changes how a portfolio should be built.
Correlation is not a property an asset has. It is a number computed from two price series over a chosen period, and choosing a different period produces a different number. Two people looking at the same asset can both be correct and completely disagree, because one measured a year and the other measured a month, and nothing in the way the figure is usually quoted reveals which.
This makes almost every claim about crypto correlation unfalsifiable as stated. Somebody asserting that it moves with technology equities can find a window where that is strongly true; somebody asserting it moves independently can find a window where that is equally true. Neither is lying and neither has said anything predictive, because the window was chosen after looking at the data.
The only version of the claim worth taking seriously specifies the period and the frequency before anybody looks at the answer. Daily returns over the last ninety days is a claim. Correlated with equities is not a claim, it is a mood, and it changes with the mood of whoever is speaking.
Ask over what window before accepting any correlation figure. Without one, the number is not a measurement of the market, it is a measurement of the speaker's chosen period.
Looking across its history rather than at any one window, crypto has spent long stretches in each of three distinguishable relationships with the rest of the financial world, and it has moved between them without warning.
For much of its early history, the asset was too small and too disconnected from institutional capital for anything happening in equities or bonds to reach it. Price moves came from adoption, from regulatory events specific to the sector, and from internal dynamics. This regime was genuinely uncorrelated, and it was a property of the market being small rather than of the asset being special.
As institutional participation grew, crypto began responding to the same conditions that move other risky assets: monetary policy expectations, liquidity in the financial system, and the general appetite for holding volatile things. In this regime it does not just correlate with technology equities, it amplifies them, falling further on bad days and rising further on good ones.
In episodes where confidence in a particular currency or banking system falls sharply, crypto has moved opposite to the local financial system rather than with it. These periods are real, they are short, and they are geographically specific, which is why they show up strongly in some data and not at all in aggregate figures.
The regimes are easier to understand once you stop looking for a narrative and look at where the money comes from. Crypto moves with other risk assets to the extent that the same participants hold both, and the strength of the relationship tracks how much of the marginal buying comes from portfolios that also hold equities.
That reframes the driver as liquidity conditions rather than sentiment. When financing is cheap and portfolios are expanding, capital flows into everything volatile at once, and correlations rise mechanically because a common factor is moving all of it. When financing tightens, the same portfolios reduce risk across the board, and the selling arrives in crypto for reasons that have nothing to do with crypto.
It also explains why the relationship is stronger during large moves than during small ones. Day to day, idiosyncratic factors dominate and the correlation looks modest. During a broad risk event, the common factor overwhelms everything else, and assets that appeared unrelated move together. That asymmetry is the single most important property of the whole subject.
The uncomfortable regularity, which holds across every asset class and not only here, is that correlations increase during stress. Positions that diversified each other in calm conditions stop doing so precisely on the days when diversification would have mattered, because the thing driving prices on those days is a common factor rather than anything specific to each holding.
In crypto this effect is more pronounced than in most markets, for a mechanical reason on top of the behavioural one. Leveraged positions across many different assets are liquidated by the same margin systems at the same moment, and forced selling in one asset pushes down the collateral value supporting positions in others. The correlation during a cascade is not a statistical observation, it is a plumbing consequence.
The practical implication is that a portfolio's diversification should be measured during stress periods rather than on average. A pair of holdings whose full-sample correlation is modest but whose correlation during the worst ten days is near one is not diversified in any sense that protects capital, and the full-sample number will never reveal it.
Measure correlation during the worst days, not on average. The average includes all the days when it did not matter.
Most attention goes to the relationship between crypto and traditional assets, and the more consequential one for anybody holding several crypto assets is the relationship among them. Across almost any window, the major assets in this sector move together far more tightly than participants assume, and the smaller ones move with the largest one more tightly still.
The reason is structural rather than accidental. Most crypto assets are quoted against the same few units, funded by the same pool of capital, held by overlapping participants, and traded on the same venues subject to the same liquidity conditions. A portfolio of ten crypto assets is frequently one position with ten labels, and the appearance of diversification comes from the names rather than from the behaviour.
The test is simple and rarely run. Take the daily returns of your holdings, compute how much of the total variation is explained by the largest asset in the sector, and see what is left. For most portfolios of this kind the answer is that very little is left, and the positions that felt like separate decisions were one decision expressed several times.
The debate between digital gold and speculative risk asset persists because both descriptions are accurate about different periods, and because each side quotes the periods that support it. There is no contradiction to resolve: the asset has behaved as both, and the switch between behaviours is what needs explaining rather than the behaviours themselves.
What determines which regime is operating appears to be the composition of the marginal buyer. When the flow is dominated by portfolios that hold it alongside equities, it behaves like the rest of those portfolios. When the flow is dominated by participants seeking to exit a specific currency or banking system, it behaves like an escape route. Neither is the asset's true nature, because it does not have one; it has a set of holders whose reasons differ.
The consequence for anybody making an allocation is that the diversification argument cannot be assumed. It has to be measured, in the current period, and rechecked, because the regime that made the argument true can end without any announcement and has done so more than once.
Three rules make the exercise informative rather than confirmatory. Fix the window before you look at the result, because choosing it afterwards guarantees you find what you expected. Use returns rather than prices, since two rising series will show a strong relationship whether or not their movements are actually linked. And compute the figure over several different windows rather than one, because a relationship that only exists in one window is not a relationship.
Then add the stress test that matters more than the average. Isolate the ten worst days for your largest holding and compute what everything else did on those days. That number tells you what your diversification is worth when it is needed, and it is routinely far worse than the full-sample figure that gets quoted in the allocation decision.
None of this argues for or against holding crypto, which is a separate question. It argues that the diversification benefit frequently claimed for it is conditional on a regime, and that regimes end. An allocation justified by low correlation should be sized as though the correlation could rise to one, because in every previous stress period it substantially did.
Within the sector, the practical version is that holding several assets is not diversification unless they have been measured against each other and against the largest one. Adding a tenth position to a portfolio whose returns are already explained by a single factor adds transaction costs and the feeling of having done something, and nothing else.
And the honest position on the wider question is that nobody knows which regime will be operating next year. The three that have occurred are describable, the switches between them were not predicted by anybody in advance, and a portfolio built on the assumption that the current one persists is a portfolio with an undeclared bet inside it.
Sometimes strongly, sometimes not at all, depending on the period measured. Both answers are supported by real data from real windows. The version of the question worth asking specifies a window and a frequency before looking at the result, because choosing the window afterwards guarantees finding whatever you expected.
Because what drives it is the composition of the marginal buyer rather than a property of the asset. When flow is dominated by portfolios that also hold equities, it moves with them. When it is dominated by participants exiting a specific currency or banking system, it moves against the local financial system. Neither is its true nature.
It has behaved that way in specific episodes, usually where confidence in a particular currency fell sharply, and it has behaved like a high-beta risk asset for longer stretches. The honest statement is that it has been both, at different times, and that nothing identifies in advance which regime is about to operate.
Usually far less than it appears. Most assets in the sector are quoted against the same units, funded by the same capital, and traded on the same venues under the same liquidity conditions. Compute how much of your portfolio's variation is explained by the largest asset in the sector and see what remains.
Because a common factor dominates on those days, and in crypto a mechanical effect adds to it: leveraged positions across many assets are liquidated by the same margin systems at the same moment, and forced selling in one reduces the collateral value supporting others. The correlation during a cascade is plumbing rather than sentiment.
Fix the window before looking at the result, use returns rather than price levels, and compute it over several windows rather than one. Then isolate the ten worst days for your largest holding and see what everything else did. That last figure is what your diversification is actually worth.
Yes, in its early years, and it was a property of the market being small rather than of the asset being special. Too little institutional capital held both crypto and traditional assets for anything happening in one to reach the other. That condition ended as participation grew, and it is unlikely to return.
Only if the position is sized as though the correlation could rise to one, because in every previous stress period it substantially did. A diversification argument that depends on a regime is a bet on that regime persisting, and the switches between regimes have not been predicted in advance by anybody.