Research · Altcoins · Strand B

The ETH-BTC correlation record

Correlation between two assets is a statement about how their returns moved together over a stated window, not a statement about one causing the other. This page sets out how the ETH-BTC relationship is measured, how the measure behaves, and the conditions under which it stops being informative.

Last reviewed 2026-09-21Source: CoinGecko historical price data; Coin Metrics community dataMethod and interpretation only. No correlation coefficient is asserted on this page.

What a correlation coefficient measures

A correlation coefficient summarises how two return series moved together across a set of observations. It is bounded between minus one and plus one, and it is calculated from returns rather than from price levels. That distinction matters more than it first appears. Two assets whose prices both trend upward over a decade will show a high correlation in levels almost regardless of what they are, because both series are dominated by the same drift. Correlating returns removes the trend and asks a narrower question: on a given day, week or month, did the two assets move in the same direction, and by how much relative to their own typical variation.

The window over which the coefficient is computed changes the answer as much as the data does. A thirty-day rolling correlation and a three-year correlation over the same pair can differ substantially, and neither is wrong. The short window responds quickly to a regime change and is noisy; the long window is stable and slow to reflect anything new. A correlation figure quoted without its window is therefore close to meaningless, which is why this page states the window before it discusses any behaviour.

There is a second measurement choice that is easy to miss. Returns can be computed at daily, weekly or monthly frequency, and the coefficient generally rises as the sampling interval lengthens. That is not evidence of a stronger relationship; it is an artefact of averaging away short-horizon noise. Two series that are only loosely related day to day can look tightly related month to month. Any comparison of correlation figures across sources has to check the sampling frequency before it compares anything.

How the measure behaves in practice

How the ETH-BTC return correlation behaves under different measurement choices, and what each behaviour does and does not indicate.
Condition	What is observed	How to read it
Daily returns, short window	The coefficient moves sharply from observation to observation	Noisy by construction; a single large day can dominate a thirty-day window
Weekly returns	Smoother than daily, still responsive within a quarter	A reasonable default for describing a regime rather than a day
Monthly returns	Higher on average than daily or weekly for the same pair	Partly an artefact of averaging; not evidence of a tighter link
During a broad market drawdown	Correlation across large assets tends to rise together	Consistent with a common market factor rather than with one asset driving the other
Across a regime change	The coefficient can shift and stay shifted for months	A rolling window is required to see it; a single full-sample figure hides it
During an asset-specific event	The two series can decouple for a period	Decoupling is informative about the event, not about a permanent change

Last reviewed 2026-09-21Source: Method description; behaviour summarised from published return seriesNo coefficient values are stated; the table describes measurement behaviour, not a result.

The most consistent feature of the record is that correlation is not constant. It rises during periods when the whole market is moving together and falls when an asset-specific story dominates. That pattern is not unique to ETH and BTC; it is a general property of assets that share a common investor base and a common liquidity cycle. When leverage is being added or removed across a market, the large assets tend to move as one, because the same participants are transacting in all of them at once.

This is why a high correlation reading is weak evidence for a specific causal story. If two assets are held by overlapping investors, quoted against the same settlement currency, and traded through the same venues, a common factor will produce co-movement whether or not either asset influences the other. The correlation coefficient cannot distinguish between those cases. It reports that the returns moved together; it says nothing about why.

Where the relationship breaks down

The relationship is least informative at the moments it is most often cited. During a sharp market-wide decline, correlations across large assets converge toward one, and a reader who takes that convergence as confirmation that the assets are fundamentally the same has read a stress artefact as a structural fact. The same convergence appears in equity indices during a crash and is not treated as evidence that every constituent company is identical.

The relationship is also unstable around changes in the relative supply and demand for each asset. A change in one network's issuance schedule, a large forced seller, or a shift in where the marginal buyer is located can move one series without moving the other. Those episodes are where the coefficient falls, and they are the episodes that carry the most information — precisely because they are the ones a common-factor explanation does not cover.

The practical conclusion is that the ETH-BTC correlation is best treated as a descriptive statistic with a stated window, useful for characterising a period and useless for predicting the next one. It is a summary of what happened, not a property of the pair that can be relied on to persist. For the underlying measures this page leans on, see Volatility Explained and Market Cap Explained.

Dataset, period, method and limitations

Dataset: Daily ETH and BTC closing prices from CoinGecko's historical data endpoints, cross-checked against Coin Metrics community data where the two overlap.
Period: From Ethereum's public trading history to the last-reviewed date stated above. Rolling windows are computed within that span; no window extends beyond the data.
Method: Returns are computed as simple period-over-period percentage changes at daily, weekly and monthly frequency. Correlation is Pearson's coefficient over a stated rolling window. No coefficient is reported on this page because the build cannot verify a specific value against a frozen dataset.
Limitations: Correlation is sensitive to the sampling frequency, the window length and the treatment of outliers. It measures linear co-movement only and can be near zero for a strong non-linear relationship. It says nothing about causation, and a high reading is equally consistent with a common factor, with reverse causation, or with coincidence over a short window.

What this page does not claim

This page does not state a correlation coefficient, because a figure quoted without a frozen dataset and a stated window cannot be checked by the reader. It does not claim that Bitcoin's price movements caused Ethereum's, or the reverse. It does not present correlation as a forecast, and it does not treat a period of high co-movement as evidence that the two assets share a fundamental value driver.

Sources and references

Every source below is named and linked. Where a page describes a method rather than a figure, the source is the specification or documentation that defines the method.

Ethereum historical data. CoinGecko, www.coingecko.com/en/coins/ethereum/historical_data: daily price and market-capitalisation series used for the return calculations.
Bitcoin historical data. CoinGecko, www.coingecko.com/en/coins/bitcoin/historical_data: the BTC leg of every pair calculation on this page.
Community data. Coin Metrics, coinmetrics.io/community-network-data/: an independent series used to cross-check the price inputs.
Correlation and dependence. NIST Engineering Statistics Handbook, www.itl.nist.gov/div898/handbook/pmc/section5/pmc542.htm: the definition of the coefficient and its sensitivity to outliers.