Key takeaways
- Bitcoin's daily correlation with the S&P 500 was 0.14 on days that opened within 5% of a record high, and 0.33 on days that opened 5% or more below one.
- Sensitivity barely moved. Bitcoin's beta to the index was 0.80 in the calm state and 0.82 in the drawdown state, so what changed is correlation, not amplification.
- Across the five S&P 500 falls of 10% or more since September 2016, a 5% bitcoin sleeve deepened the peak-to-trough loss in three and softened it in two.
- The 2022 fall cost the most. A 25.4% equity loss became 27.1% with a 5% sleeve and 28.8% with a 10% sleeve.
- Bitcoin's own worst recent stretch, 51.7% from 6 October 2025 to 30 June 2026, ran alongside an S&P 500 gain of 11.3%.
Bitcoin's correlation with the S&P 500 more than doubled once shares were already falling
Does holding crypto spread a portfolio's risk, or concentrate it? On ten years of daily closes the answer flips depending on what equities were doing at the moment you asked. Bitcoin's daily correlation with the S&P 500 was 0.14 on days that opened with the index within 5% of its record high. It was 0.33 on days that opened 5% or more below that high, and 0.38 on days that opened 10% or more below it. The diversification was strongest in the calm stretches and weakest in the falls.
Both series come from FRED. One is Coinbase's daily bitcoin close in US dollars, series CBBTCUSD. The other is the S&P 500 daily index close, series SP500. They share 2,511 trading days between 6 September 2016 and 2 September 2026, which gives 2,510 paired daily returns. FRED holds ten years of daily history for the S&P 500 series, and that is what fixes the start of the window.
The state test uses the previous close, not the current one. A day is classified by how far the index had already fallen from its running peak when the day began, so nothing about that day's move is used to sort it. That matters. Sorting days by their own outcome manufactures the result you were looking for.
Over the whole sample the correlation was 0.22. That single number hides the split. The drawdown state covered 826 of the 2,510 days at the 5% threshold and 429 at the 10% threshold, so the high-correlation regime isn't a handful of outliers. It's about a third of the decade.
Two other cuts point the same way. Conditioning on the previous day's VIX instead of on the drawdown, bitcoin's correlation with the index was 0.11 across the 1,730 days that followed a VIX close under 20, and 0.42 across the 153 days that followed a close above 30. Ether behaved almost identically to bitcoin: 0.17 in the calm state and 0.33 in the 5% drawdown state, on the same 2,511 days. The pattern isn't a bitcoin quirk. What ether does not share is bitcoin's risk profile, and the gap between ether volatility and bitcoin's held in every start window we have tested.
Crypto diversification changed through correlation, not through the size of the move
Here's where crypto diversification gets misdescribed. Bitcoin's beta to the S&P 500 was 0.80 in the calm state and 0.82 in the 5% drawdown state. Beta is the average move in bitcoin per 1% move in the index. On that measure, almost nothing happened.
Correlation answers a different question: how much of bitcoin's movement the index accounted for. Squaring the two correlations gives roughly 2% of bitcoin's daily variation in the calm state and roughly 11% in the drawdown state. In quiet markets crypto's own news dominates, and the equity link is buried under it. In falling markets the equity signal surfaces. Bitcoin didn't start moving more per unit of equity move. It started moving with equities more of the time.
That distinction has a practical edge. A sleeve whose beta is stable but whose conditional correlation rises doesn't blow up a portfolio. It quietly stops offsetting anything, at the point in the cycle when offsetting is the entire reason it's there. The same fragility shows up across asset classes, and it's the reason correlation instability undoes more diversification plans than any single bad asset does.
One more cut is worth showing, with a caveat attached. Splitting days by the sign of the same-day equity return, bitcoin's downside beta was 1.14 on the 1,135 days the index fell and 0.54 on the 1,374 days it rose. That looks like a clean asymmetry, and it partly is. But sorting days by the same-day equity move also sorts them by whatever drove both series that day, so the gap overstates a true downside beta. It describes what a holder experienced. It isn't a clean estimate of the underlying sensitivity.
None of this is unique to crypto. Longin and Solnik, writing in the Journal of Finance in 2001 on international equity markets, concluded that "correlation is not related to market volatility per se but to the market trend. Correlation increases in bear markets, but not in bull markets." Crypto is a recent instance of an old finding.
Five equity falls of 10% or more, and crypto worsened three of them
Regime statistics are averages. A holder lives through specific episodes, so here are all five S&P 500 falls of 10% or more inside the sample, measured peak close to trough close, with bitcoin's return over the identical window.
- 26 January to 8 February 2018. S&P 500 down 10.2%, bitcoin down 25.9%, ether down 22.3%. Nine trading days.
- 20 September to 24 December 2018. S&P 500 down 19.8%, bitcoin down 37.9%, ether down 38.0%.
- 19 February to 23 March 2020. S&P 500 down 33.9%, bitcoin down 31.4%, ether down 46.6%.
- 3 January to 12 October 2022. S&P 500 down 25.4%, bitcoin down 58.8%, ether down 65.7%.
- 19 February to 8 April 2025. S&P 500 down 18.9%, bitcoin down 16.0%, ether down 39.0%.
Bitcoin fell less than the index in two of the five, and both were the fast ones. In the 2020 crash it lost 31.4% against the index's 33.9%, despite dropping 37.4% in the single session of 12 March 2020, when the index fell 9.5%. It recovered part of that inside the window. In the 2025 fall it lost 16.0% against 18.9%.
The within-episode correlations track the speed. The 2020 window ran at 0.61 and the 2022 window at 0.57, both above the full-sample 0.22. The autumn 2018 window ran at 0.06 and the 2025 window at minus 0.03. Crypto tracked equities most closely in the two episodes where the whole market was repricing at once, and barely at all in the two where equities fell for reasons of their own.
Crypto sleeve size against outcome, inside each of the five falls
The table below applies the episode returns to a portfolio that held the index plus a fixed bitcoin weight at the equity peak, then rode the drawdown without trading. It's the peak-to-trough loss a holder recorded at each crypto sleeve size.
- Jan to Feb 2018: 10.2% at no sleeve, 10.5% at 2%, 10.9% at 5%, 11.7% at 10%.
- Sep to Dec 2018: 19.8%, 20.1%, 20.7%, 21.6%.
- Feb to Mar 2020: 33.9%, 33.9%, 33.8%, 33.7%.
- Jan to Oct 2022: 25.4%, 26.1%, 27.1%, 28.8%.
- Feb to Apr 2025: 18.9%, 18.8%, 18.8%, 18.6%.
The worst outcome in this data is 2022, where a 10% sleeve added 3.3 percentage points to an already bad year. The best is a rounding error in the other direction. At the sizes people actually hold, the sleeve moved the depth of an equity drawdown by less than a percentage point in four of the five episodes. That is the honest headline, and it cuts against the alarmist version of the argument as much as against the optimistic one.
Run the same blend daily rebalanced across the entire ten years and the pattern holds. The index alone compounded at 13.42% a year with a worst drawdown of 33.9%. At a 5% weight the return was 16.64% and the worst drawdown 33.7%. At 10% it was 19.83% and 33.5%. Only at a 20% weight did the worst drawdown rise, to 36.0%, and it moved to a different date: the October 2022 trough rather than the March 2020 one. What a sleeve did to return was large. What it did to equity drawdown depth was small until it stopped being a sleeve. The same asymmetry appears in the longer-run test of a bitcoin allocation inside a 60/40, where the start date does more work than the weight.
Most of crypto's own damage arrived when equities were fine
The framing of this topic assumes crypto's bad days are equity bad days. Mostly they aren't. On bitcoin's 20 worst days in the sample it fell 15.84% on average, while the index fell 1.73% on the same days. The index was down on 15 of those 20 days, so the link isn't nothing. It's small relative to what bitcoin was doing.
The clearest case is recent. Bitcoin fell 51.7% between 6 October 2025 and 30 June 2026. Over that same stretch the S&P 500 rose 11.3%. A holder lost roughly half of the sleeve while the rest of the portfolio was making money, which is a different problem from the one this article set out to test and arguably the more common one.
Scale explains part of the asymmetry. The Bank of England's March 2022 Financial Stability in Focus put it plainly: "Bitcoin returns are three times as volatile as the S&P 500." On this sample the gap is wider still, with annualised daily volatility of 66.6% for bitcoin against 18.1% for the index. When one series moves three or four times as much as the other, a correlation of 0.33 still leaves most of the crypto move unexplained by equities.
The counter-case: 0.33 is a low number, and the sample is short
The strongest objection to the argument above is that 0.33 is not a high correlation. Assets that genuinely fail as diversifiers in a crisis go to 0.8 and beyond. High-yield credit and small-cap equity do that. Bitcoin, on this data, does not. Someone holding a 5% sleeve for its long-run return, and expecting no help in a crash, gets exactly what the data describes. The evidence here doesn't refute that position. It refutes the claim that crypto is an uncorrelated asset, which is a different and much stronger claim. Crypto diversification of the weak sort is still diversification.
The second objection is the sample. Ten years is one full crypto cycle and change, containing one pandemic crash, one rate shock and two smaller falls. Five episodes is not a distribution. The IMF's 2022 Global Financial Stability Note on crypto and equity spillovers found the same directional result on 2017 to 2021 data, reporting that "spillovers tend to increase during episodes of market stress" and putting bitcoin's return correlation with the S&P 500's sub-sectors in a range of 0.27 to 0.38. Two overlapping samples agreeing is weaker evidence than it looks.
The third is measurement. Coinbase's daily figure is struck at 5 PM Pacific, several hours after the New York equity close, so a same-day pairing gives bitcoin information the index close didn't have. That should inflate the measured link. Testing for it, bitcoin's return against the next day's index return was 0.003, and the index against the next day's bitcoin return was minus 0.025. Neither is meaningful, which suggests the timing overlap isn't manufacturing the result, but it doesn't eliminate the concern.
The fourth is that regimes are visible only afterwards. A rolling 90-day correlation on this data ran from minus 0.34 in August 2019 to 0.62 in July 2022, and sat at 0.30 on 2 September 2026. It was above 0.3 in 31.9% of windows and below zero in 20.5% of them. Anyone sizing a sleeve to a correlation estimate is using a number that has swung by nearly a full point inside one decade. Broader evidence on diversification in a crisis shows the same estimate problem across every asset class that has ever been sold as a hedge.
UK readers have a further constraint that no correlation table captures. The FCA states that "Cryptoassets remain high risk" and that "consumers should be prepared to lose all of their money if they buy cryptoassets", with no Financial Services Compensation Scheme cover if things go wrong. A conditional correlation figure describes co-movement. It says nothing about whether the asset survives.
What would change this conclusion
The result rests on one mechanism: crypto trades as a long-duration risk asset, so it reprices when the price of risk itself reprices. Three things would break that.
The first is a fall driven by something other than the price of risk. Every episode here except autumn 2018 was a broad repricing. An equity drawdown caused by a sector, an earnings collapse or a domestic policy shock would leave crypto free to do something else, which is roughly what the 2025 window's correlation of minus 0.03 shows.
The second is a change in who holds it. The 0.14 calm-state figure comes largely from years when crypto's marginal buyer was not the marginal buyer of equities. As those holder bases converge, the calm-state number has room to rise toward the drawdown-state number, and the conditional gap this article is built on would shrink from the wrong end.
The third is a crypto-specific failure large enough to dominate. The 51.7% fall into June 2026 happened while equities rose. A repeat of that during an equity drawdown would produce a much worse joint outcome than a correlation of 0.33 implies, because correlation measures co-movement in ordinary days and says little about a simultaneous tail.
The number to watch isn't the headline correlation. It's the gap between the calm-state and drawdown-state readings. It stood at 0.19 points on 2 September 2026, at 0.14 against 0.33. A gap that closes toward zero would mean crypto had become an ordinary risk asset, priced off the same factor as everything else, and the crypto diversification argument would be over on its own evidence rather than on anyone's opinion.