Key takeaways
- Weekly bitcoin and S&P 500 returns correlated 0.19 across the 611 weeks from 12 December 2014 to 2 September 2026, on Coinbase and Cboe closing prices.
- The 52-week rolling correlation ran from -0.226 in December 2019 to 0.546 in July 2026. It sat below 0.2 in 53.2% of the 560 windows.
- In the 152 weeks that opened with the highest VIX, correlation was 0.248. In the 152 calmest weeks it was 0.115, so the link tightens exactly when it costs most.
- Across the 31 worst weeks for the S&P 500, the index averaged -5.45% and so did bitcoin. Bitcoin fell in 65% of them.
- A 5% bitcoin sleeve rebalanced weekly moved 11-year portfolio volatility from 16.77% to 16.84%, and the worst drawdown from -31.8% to -31.9%.
Bitcoin equity correlation is 0.19 over 11 years, and the average is the least useful number in the series
You want to know whether bitcoin still diversifies a stock portfolio, or whether it has quietly become another way to own the same risk. Here is the arithmetic, computed from two fetched price series.
Over the 611 weeks from 12 December 2014 to 2 September 2026, weekly bitcoin returns and weekly S&P 500 returns correlated 0.19. That is low. A reading of 1.0 would mean the two moved in lockstep, and 0 would mean no linear relationship at all.
It is also an average of 11 years, and it describes almost none of them. The series that produced 0.19 contains a year at -0.09 and a stretch above 0.5, and the difference between those regimes is the whole question.
The prices are Coinbase's daily bitcoin close, published by the St. Louis Fed as series CBBTCUSD and timestamped "All data is as of 5 PM PST", and Cboe's own daily history of the S&P 500. Returns are weekly logarithmic changes, taken only on days both markets printed a price, so bitcoin's weekend trading is excluded rather than averaged in.
The 52-week rolling correlation has ranged from -0.226 to 0.546
Roll a 52-week window through the weekly returns and you get 560 overlapping estimates. The lowest was -0.226, in the window ending 13 December 2019. The highest was 0.546, ending 10 July 2026. The median was 0.167, and the window ending 2 September 2026 read 0.376.
That range is the finding. In 19.3% of the windows the correlation was negative, and in 53.2% it was below 0.2. In 3.2% it was above 0.5, and those windows aren't scattered: they cluster in two episodes, autumn 2022 and the summer of 2026.
The chart plots the same data annually, correlating weekly returns within each calendar year. Bitcoin's relationship with equities was -0.013 in 2016 and -0.094 in 2019. It jumped to 0.305 in 2020, reached 0.429 in 2022, fell back to 0.130 in 2023, then rose to 0.369 in 2025 and 0.243 through 2 September 2026.
The IMF measured the same break from a different angle. Tara Iyer's January 2022 note for the Fund compared January 2017 to December 2019 with January 2020 to November 2021, and found the correlation between bitcoin price volatility and S&P 500 volatility "has increased more than four-fold", with bitcoin's contribution to the variation in S&P 500 volatility up "about 16 percentage points". Two different datasets, two different methods, the same turn.
Correlation rises with the VIX, which is the regime where you'd want it not to
Sort the 610 weeks by the VIX level at the start of each one and split them into quartiles. The VIX is Cboe's own gauge, and its methodology defines it plainly: "The VIX Index measures 30-day expected volatility of the S&P 500 Index."
In the calmest quartile, the 152 weeks that opened with the VIX at or below 13.48, bitcoin and the S&P 500 correlated 0.115. In the most fearful quartile, the 152 weeks opening at or above 20.97, they correlated 0.248. The middle half came in at 0.182.
That gap survives a crude robustness test. Drop any single week from either quartile and the stressed reading stays between 0.174 and 0.302, while the calm reading stays between 0.062 and 0.138. No individual week is carrying it.
So the correlation roughly doubles between calm and stress. That is why correlation instability matters more than the headline coefficient. A number measured across all weather tells you what happens on the average day, and nobody's portfolio is tested on the average day.
In the worst 5% of equity weeks, bitcoin fell exactly as much
Take the 31 weeks with the worst S&P 500 returns in the sample. The index averaged -5.45% across them. Bitcoin averaged -5.45% too, and it fell in 65% of them.
Two stress windows show the mechanism at daily frequency. From 19 February 2020, the S&P 500 fell 33.9% to its trough on 23 March 2020. Bitcoin fell 48.1% to its own trough on 12 March 2020, eleven days earlier. Through 2022, from the first trading day of January, the index was down 25.4% at its 12 October low; bitcoin was down 66.1% by 21 November.
Neither episode shows bitcoin cushioning an equity fall. Both show it amplifying one, on a delay of days. That is a different claim from "correlation is high", and it is the claim that matters for anyone sizing a sleeve, because drawdowns are additive in a way that average correlation is not.
What a bitcoin sleeve did to portfolio volatility, at five sizes
Correlation only earns its keep once you attach it to a weight. Here is the same 611 weeks run as a portfolio: a bitcoin sleeve alongside the S&P 500, rebalanced back to its target every week.
| Bitcoin sleeve | Annualised volatility | Worst drawdown |
|---|---|---|
| 0% | 16.77% | -31.8% |
| 1% | 16.73% | -31.8% |
| 3% | 16.73% | -31.9% |
| 5% | 16.84% | -31.9% |
| 10% | 17.54% | -32.0% |
| 20% | 20.51% | -35.2% |
Read the top rows and the risk cost of a small sleeve looks like a rounding error. A 5% weight added 0.07 points of annualised volatility over 11 years. The 1% and 3% rows are marginally below the equity-only line, which is what a 0.19 correlation is supposed to do.
Read the bottom row and the arithmetic reasserts itself. At 20% the sleeve added 3.74 points of volatility and 3.4 points of drawdown. The reason isn't correlation at all. It's the volatility ratio: bitcoin's weekly returns annualise to 66.0% against 16.8% for the index, so a low correlation is being multiplied by an asset that moves roughly four times as much.
The last five years make the same point with less flattering data. From 3 September 2021 to 2 September 2026 the correlation was 0.283, and a 5% sleeve took volatility from 16.25% to 16.35% and the worst drawdown from -24.8% to -26.7%. The drawdown cost was 1.9 points there against 0.1 across the full sample, while the volatility cost barely moved. What this table deliberately leaves out is return, which depends almost entirely on the start date and is covered separately in our work on bitcoin allocation inside a balanced portfolio.
For a UK reader there is also a regulatory ceiling on the same question. The FCA proposed in its 2022 consultation "to classify cryptoassets as 'Restricted Mass Market Investments'", and confirmed the rules in June 2023. Its conduct rules, in force from 8 October 2023, explain that the restricted investor statement's reference to "not investing more than 10% of their net assets" applies to "the retail client's aggregate investment across all types of restricted mass market investment".
The strongest case against reading this as a diversification failure
A fair reader can look at everything above and conclude the opposite of alarm, and the numbers support them.
A stress correlation of 0.248 is not a broken hedge. It is a long way from lockstep, and it means bitcoin's bad weeks and the market's bad weeks line up loosely rather than mechanically. In 19.3% of rolling windows the correlation was negative outright. And the portfolio evidence is blunt: at the sleeve sizes people actually hold, the measured cost was 0.07 points of volatility over 11 years and 0.10 over the last five.
The honest referee's verdict is that both readings are describing the same table. Correlation has risen, and it rises further under stress, but from a base low enough that a 1% to 5% sleeve barely registers in portfolio volatility. The thing that registers is bitcoin's own 66.0% volatility, which is a sizing question rather than a correlation question. The rolling correlation matters most for the reader who assumed a negative number and sized accordingly.
Left alone, a 5% sleeve became 74.5% of the portfolio
Every figure in that table assumes weekly rebalancing, and that assumption is doing more work than the correlation is.
Run the same 5% sleeve with no rebalancing at all and it ends the sample at 74.5% of the portfolio, having peaked at 84.4%. Volatility rises from 16.84% to 38.11% and the worst drawdown goes from -31.9% to -62.8%. Rebalance once a year instead of weekly and the sleeve still reaches 47.0% of the portfolio at its peak, because a single year can carry it there.
That is why a correlation study and a crypto rebalancing study answer different questions. The correlation tells you what a 5% sleeve does while it stays a 5% sleeve. Whether it stays one is a separate mechanism, and on this data it doesn't.
What this bitcoin equity correlation data cannot tell you
Start with the sample. It is 611 weeks. That is one asset, one exchange's price, one equity index, and one long expansion punctuated by two bear markets. Correlations estimated on 52 weeks carry real estimation error, and overlapping windows make consecutive readings look steadier than independent ones would.
The bigger caveat is the one most correlation studies skip. These are Friday-to-Friday returns, and the day you choose matters more than most of the regimes described above. Measure the same 11 years Thursday to Thursday and the full-sample correlation is 0.284 rather than 0.19, and the highest 52-week window is 0.669 rather than 0.546. The direction of every finding here survives that switch. The decimals do not, and any single quoted correlation should be read with that in mind.
The price series has its own limitations. CBBTCUSD is Coinbase's 5 PM PST print, not a volume-weighted composite across venues, and bitcoin trades continuously while the S&P 500 does not. Aligning them on common trading days is the cleanest available choice, and it still discards every weekend move, some of which were large. Frequency matters too: across the 61 trading days from February to April 2020, daily returns correlated 0.552, against 0.195 for daily returns across the whole sample.
It is also a US index. A sterling investor holds the same bitcoin exposure with a currency leg attached, and this study measures neither the pound nor a UK equity benchmark. And the whole exercise is a backtest of the one cryptoasset that survived; the correlation record of the ones that didn't isn't in the data at all.
Finally, a backtest is not what investors experienced. The BIS studied crypto app downloads and put it this way: "The median investor would have lost $431 by December 2022, corresponding to almost half of their total $900 in funds invested since downloading the app." Correlation arithmetic describes a disciplined rebalanced sleeve. Entry timing did most of the damage in practice, and it doesn't appear in any correlation figure.
What would change the conclusion
If the rolling correlation stayed above 0.5 for a full cycle, rather than spiking there in autumn 2022 and mid-2026 and reverting, the diversification reading would have to go. Two clusters covering 3.2% of windows is an episode. A persistent 0.5 would be a regime.
If bitcoin's volatility kept falling toward the index's, the sizing arithmetic changes more than any correlation shift could. At 66.0% against 16.8%, the volatility ratio dominates. Halve it and a 10% sleeve starts behaving like today's 5%.
If a stress episode arrives with bitcoin held mostly by long-horizon owners, the 0.248 high-VIX reading could break in either direction, and this sample contains no such episode to test it against.
LedgerTouch computes pairwise correlations across a portfolio's holdings from their price histories, which is one way to see a regime change while it's happening rather than after it. But the number worth watching isn't the 11-year average, and it isn't any single coefficient. It's the current 52-week reading against its own history: 0.376 as of 2 September 2026, against a median of 0.167 and a peak of 0.546.