The spreadsheet had ten tokens in it, and the friend who sent it was genuinely proud of the spread. Two layer ones, a DEX token, an oracle, a liquid staking token, a gaming coin, a meme position he described as ironic. Different sectors, different theses, different founders. He wanted to know if ten was enough to count as properly diversified. So I pulled ninety days of daily closes, ran the pairwise correlations, and nearly every number in the matrix came back above 0.8. He had done the research work of ten separate investments and ended up holding roughly one.
This result is boringly normal. If you hold a basket of altcoins and have never computed the correlations, I would bet on your matrix looking about the same. The sector labels describe what the projects do. They tell you almost nothing about what the tokens do on the days the whole market drops, and that is the only sense of diversification your account actually cares about.
Correlations rise exactly when you need them to fall
In calm markets, large-cap altcoins typically correlate with bitcoin at a level high enough to raise an eyebrow but low enough that you can convince yourself the spread is doing something. In drawdowns, the pairwise numbers historically converge toward 0.8 to 0.9, sometimes higher. Traditional markets have a version of this too, the old line about correlations going to one in a crisis, but crypto has structural reasons for doing it harder and faster. The timing is the cruel part of the arrangement. Diversification is a tool you buy for the bad days, and this is an asset class where it reliably stops working right as the bad days start.
Nearly everything trades against BTC and ETH pairs, so those two assets are the liquidity backbone for the whole asset class. The marginal buyer of an oracle token and the marginal buyer of a gaming token is largely the same pool of risk-seeking capital, and when that capital leaves, it leaves everything at once. Leverage then makes the whole thing mechanical. A liquidation cascade in one corner forces selling across entire portfolios, market makers widen or pull quotes everywhere at the same time, and the idiosyncratic stories that made your ten tokens feel different get flattened by one flow of forced selling. To a first approximation, an altcoin basket is a levered bitcoin position with extra ways to lose.
The math is what makes this fatal to the ten-token idea. The volatility reduction you get from adding positions depends almost entirely on how correlated they are. With genuinely independent assets, spreading across ten names cuts portfolio volatility substantially. With pairwise correlations around 0.9, ten assets behave almost exactly like one, and what you have gained is trading fees, more things to monitor, and a false sense of safety. The historical record is consistent on this. In each of the major crypto bear markets to date, most of the large alts drew down harder than bitcoin itself, together, on the same weeks.
Run the matrix on your own holdings
Checking this takes maybe twenty minutes and needs nothing more exotic than a spreadsheet. Here is the version I actually run.
- Pull daily closing prices for every token you hold, going back at least 180 days where the token has existed that long. Any of the free market data sites will let you export this.
- Convert prices to daily returns. Simple percent change is fine, log returns if you want to be tidy about it.
- Compute the correlation of returns for every pair of holdings. In a spreadsheet that is the CORREL function on two columns. In Python it is one call to .corr() on a dataframe of returns. You end up with a grid, ten by ten if you hold ten tokens.
- Then do the step almost everyone skips. Filter to only the days bitcoin fell by more than some threshold, say 3 percent, and recompute the matrix on that subset. This is your stress correlation, and it is the number that describes your actual risk.
- Compare the two grids. The gap between them is the gap between how diversified you feel and how diversified you are.
Reading the output is straightforward. Anything above roughly 0.8 in the stress matrix means those two tokens are one position for risk purposes. In most alt portfolios the entire matrix clears that bar, which means the honest way to size the basket is as a single trade. If your personal rule is that no one position gets more than a fixed slice of your capital, that rule should apply to the whole correlated cluster. Most people who run this exercise discover they are holding several times their intended maximum in what is functionally one bet, which explains why the drawdowns keep surprising them.
Two failure modes to watch for. The first is computing correlation over a long, quiet window, seeing a comfortable number, and stopping there. Correlation is regime dependent, and the calm-market figure is precisely the one that stops describing reality when you need it to. The second is young tokens. A coin that launched a few months ago and shows low correlation to everything may simply not have lived through a stress event yet. Absence of evidence here is mostly absence of history.
What an uncorrelated addition actually looks like
Inside crypto, the honest answer is that there is very little on offer. Bitcoin is the lowest-beta asset in the class, so rotating part of an alt basket into BTC reduces risk even though it feels like doing less. Stablecoins are the one real diversifier available inside a crypto account, because a cash position has a correlation of roughly zero to everything by construction, and it can typically earn some yield while it sits there. People resist this because holding cash feels like not playing, but a half-cash book has roughly half the drawdown of a fully invested one, and that is arithmetic rather than genius. Unglamorous, and it works, which is how most things in portfolio construction seem to go.
Outside crypto, short-duration government debt and gold have historically shown low correlation to crypto drawdowns, while broad equities sit somewhere in the middle, since risk-off episodes tend to hit stocks and coins together. The more interesting move for an active person is diversifying across return streams rather than only across assets. A simple trend rule that moves a position to cash when price drops below a long moving average produces a return stream with genuinely different behavior from buy-and-hold, even applied to the same coin, because it is flat or defensive during exactly the stretches when the correlations converge. Two strategies on one asset can end up less correlated with each other than ten assets held the same way.
None of this says you have to sell anything. Ten tokens is a perfectly fine way to express the view that crypto goes up, as long as you know that is the position you actually hold and you size it like the single leveraged bet it is. Run the matrix once, rerun it after any big market move, and pay attention to the down-day version rather than the flattering full-sample one. The spreadsheet takes twenty minutes. Learning the same thing from a drawdown takes longer and costs considerably more.