With how much the number itself moves
The matrix reports the correlation of daily returns between the major cryptocurrencies over whichever window you select. That part is standard, and on its own it invites a mistake: reading a single number as if it described a stable relationship. Split the same window into quarters and the figure for a given pair often moves substantially, which means a portfolio built on the average was designed for a relationship that did not hold for much of the period.
So the second table shows that spread directly. It also reports what each coin did on the days Bitcoin fell more than three percent — because diversification is only tested when things fall, and an all-period average hides exactly those days. That column deliberately gives a median move rather than a "crash correlation": restricting a sample to days when one variable moved a lot distorts correlation mechanically, so such a figure can rise or fall for reasons unrelated to whether diversification worked. How much each coin actually dropped has no such artefact.
The broader conclusion for crypto is unflattering. Almost every pair sits well above zero, and the low readings are low by degree rather than in kind. A basket of large-cap coins is closer to one position held in several proportions than to a diversified portfolio, and the correlations that matter most tend to tighten precisely when you would want them not to.
⚠️ Not investment advice. Correlation measures past linear co-movement over a chosen window and carries no forecast. All figures come from Binance daily closes. All decisions and risks are your own.
The linear co-movement of daily returns between the major cryptocurrencies over the window you select. A value of 1 means the two moved together every day, 0 means no linear relationship, and negative values mean they tended to move in opposite directions.
Because correlation is a property of a pair over a window, not a fixed property of the pair. Splitting the same window into quarters often moves the figure substantially, which means a portfolio designed around a single average was built for a relationship that did not hold for much of the period.
Much less than the numbers suggest at a glance. Almost every large-cap pair sits well above zero, so the low readings are low by degree rather than in kind. A basket of them behaves closer to one position held in several proportions than to a diversified portfolio.
Because restricting a sample to days when one variable moved a lot distorts correlation mechanically, through range truncation. A "crash correlation" can therefore rise or fall for reasons unrelated to diversification. How far each coin actually fell on those days has no such artefact and answers the question directly.
It compares median moves on the days Bitcoin fell more than three percent. A value of 1.5 means the coin typically fell about half again as much as Bitcoin did on those days, which is the practical form of the question people are asking when they check correlation.
No. It describes how two assets moved together in the past over one window, and this page shows how unstable even that description is. It carries no forecast about direction for either asset.