Bitcoin vs Stocks and Gold: What Fifteen Years of Data Actually Show

On the 3,757 trading days our three price series share — 18 August 2011 to 29 July 2026 — a dollar in bitcoin became $5,864, a dollar in the S&P 500 with dividends reinvested became $8.43, and a dollar in gold became $2.19. That is the whole case for bitcoin in one line, and it is also the least useful line in this article.

The useful part is what sits underneath it. A return means nothing until you know what it cost in risk, so put the risk in the denominator: divide an asset’s annual return by its volatility — the typical size of its price swings — and the answer says how much return each unit of turbulence bought. It is a ratio, not a percentage, and higher is better. On that measure bitcoin’s edge over the index nearly vanishes: 0.96 against 0.90, and that from the most favourable start date our data contains. Move the start to January 2018 and the ranking inverts — bitcoin 0.29 against the index’s 0.73 and gold’s 0.85, which is to say bitcoin’s risk went unpaid while both benchmarks earned theirs. Three claims usually attached to bitcoin — digital gold, inflation hedge, uncorrelated asset — survive this data to very different degrees. One is half true, one is contradicted where it matters most, and one we cannot test at all, which is itself worth saying out loud.

If you read nothing else: on this record bitcoin has been a return engine, not a shock absorber. Kept small and topped back up to a constant 5% beside 95% in the index, it lifted the annualised return of the pair from 15.3% to 19.7% while volatility rose only from 17.0% to 17.3% — the full table is further down. On its own, though, it did not pay for its risk from either of the later starts we measured, January 2018 and January 2021, and it fell on ten of the index’s twenty worst days — so size it as the volatile sleeve you could sit through an 85% fall in, not as insurance for everything else you own.

How three different calendars are made to line up

Bitcoin trades 365 days a year; the S&P and the London gold auction do not. Everything below is computed on the trading days of the S&P, each series taking its last known close on or before that date — so bitcoin’s weekends never enter a denominator, and Friday’s index price is never stretched across a Saturday.

• What the numbers are. Bitcoin: our own daily UTC closes. Stocks: the S&P 500 total return index, dividends included — using the price index instead would hand bitcoin an unearned head start. Gold: the LBMA afternoon auction, dollars per ounce. See our methodology.
• What they leave out. Fees, spreads, taxes, storage, currency conversion and inflation. This is a comparison of three price histories, not of three portfolios.
• What this article is not. A forecast or a recommendation. Fifteen years is a small sample no matter how many daily rows it holds.

The live headline table — total return, drawdown, how often bitcoin beat each benchmark over rolling windows — is the comparison tool, and it recomputes daily. This article deliberately does not repeat it. It asks what a table cannot answer: how much of the result was the start date, what the return cost in risk, and whether bitcoin was somewhere else on the days it needed to be.

The return is mostly a statement about a date

A 5,864x is arithmetic performed on a $10.90 starting price. Shift the entry point and the answer moves more than most people expect. Below is the annualised return — the steady yearly rate that would have produced the same final result — to 29 July 2026 from the first trading day of each year: same three assets, same finish line, fourteen different beginnings.

Bitcoin’s annualised result spans 114 percentage points across those fourteen starts, from +90.6% a year to −23.3% a year; the S&P’s spans 10.3 points (11.3% to 21.6%) and gold’s 23.6 (6.5% to 30.1%). Two consequences follow, pointing in opposite directions.

First, bitcoin finished ahead of the index from ten of the fourteen entry years, and ahead of gold from ten as well. Second, it trailed both from a 2021, 2022, 2024 or 2025 start, and from a 2022 start gold more than doubled it (19.0% a year against 7.2%). Fourteen overlapping starts are not fourteen independent experiments — they mostly re-measure the same two cycles — but they are enough to show that “bitcoin outperformed” is a sentence with a hidden date in it.

What the return cost

Return without risk is half a number — bitcoin’s 78.7% a year was bought with swings nearly five times the width of the index’s. Here is the other half over the full shared window:

Measured 18 Aug 2011⁠–⁠29 Jul 2026	Bitcoin	S&P 500 TR	Gold
Annualised return	78.7%	15.3%	5.4%
Annualised volatility	81.6%	17.0%	16.7%
Return ÷ volatility (higher is better)	0.96	0.90	0.32
Deepest fall from a peak	−84.9%	−33.8%	−44.6%
…and days from that peak back to it	1,177	173	3,244
Share of days 20%+ below own peak	73.7%	1.5%	46.1%

That −84.9% row is easy to read past, so here it is in money. $10,000 put into bitcoin at its December 2013 peak was worth $1,514 at the January 2015 low, and did not see $10,000 again until 23 February 2017 — 1,177 days, more than three years spent below water. The same $10,000 in the index at its February 2020 peak bottomed at $6,621 and was whole again in 173 days.

Three rows deserve more attention than the return at the top. First, return ÷ volatility, where a bigger number means more return for each unit of risk taken: bitcoin’s 0.96 against the index’s 0.90 says its extra return came with almost exactly proportional extra risk — and that is the reading from the most favourable start date our data contains. Second, gold over these particular fifteen years was neither high-returning nor safe: $10,000 at its September 2011 peak was worth $5,538 at the December 2015 low and took 3,244 days, close to nine years, to come back. Third, bitcoin spent 73.7% of all days at least a fifth below its own record, against 1.5% for the index; what that feels like from the inside is our drawdown census.

Now do the same ratio from four different start dates, which is where the comfortable version of this comparison falls apart:

Bitcoin’s advantage in return per unit of risk is not a property of bitcoin; it is a property of the early years being in the window. It survives a January 2015 start — 0.87 against the index’s 0.76 and gold’s 0.71 — and then disappears. From a January 2018 start the index delivered 0.73 and gold 0.85 against bitcoin’s 0.29; from January 2021, 0.88 and 0.80 against 0.23. Keep the direction in mind: 0.29 is not a modest positive result, it is well under half of what the index returned for each unit of risk it asked of you. Both later windows still contain a full bitcoin cycle — a record high, a fall of more than half, a new record after it — so this is not a bear market cherry-picked. Plainly: whether bitcoin paid you for its risk depends on when you started, and from the two later starts we measured it did not. Four overlapping windows, not a projection of the next one.

Claim one: “uncorrelated”

Across all 3,756 daily returns in the window, the correlation between bitcoin and the S&P is 0.165, and between bitcoin and gold 0.093. Correlation runs from −1 to +1: zero means the two move with no relation to each other, +1 means they move in lockstep, −1 means one rises exactly as the other falls. So yes, quoted alone those two numbers sound like independence. They are also averages over fifteen years in which the relationship was not one thing.

That chart is a rolling window: on every trading day we recompute the correlation over the previous year of returns, so the line shows how the relationship drifted instead of flattening it into one number. The trailing one-year correlation with equities sat near zero for most of the years to 2019 — its minimum, −0.18, came on 16 December 2019 — then stepped up in 2020 and has spent much of the time since above 0.3. Across all 3,505 rolling readings, 31.4% are above 0.3 and 26.8% are below zero; the median — the middle reading, with half above and half below — is 0.13, the maximum 0.57 (10 February 2023) and the last one, on 29 July 2026, is 0.51. Those readings overlap almost completely, since each one shifts the year-long window by a single day, so treat 3,505 as one long trace rather than 3,505 separate observations. Whether that shift is permanent, this chart cannot say. What it can say is that “bitcoin is uncorrelated” describes the first half of our data much better than the second.

An average is not what a holder wants to know anyway. The question is narrower and harsher: on the days the rest of the portfolio is falling apart, where is bitcoin? Here are the twenty worst single days the index had in this window.

Bitcoin closed lower on 10 of the 20 and fell further than the index itself on 8. Its median move — the middle one of the twenty — was −1.2%, which looks genuinely unrelated; the mean, the plain average, was −5.9%. Most of the distance between those two numbers is one day, 12 March 2020, when the S&P lost 9.5% and bitcoin lost 39.5% — but not all of it: leave that day out and the mean over the other nineteen is still −4.1%, more than three times the median, because 5 February 2018 (−21.8%) and 13 June 2022 (−22.7%) are in there too. Gold, left off the chart because its moves are invisible at that scale, was down on 12 of the 20; its mean move was −0.5% and its worst among them −5.0%: not a shield on those days, simply somewhere else on the scale.

Twenty days is a thin slice, so widen it. Our window holds 120 days on which the index fell 2% or more; across all of them bitcoin’s average move was −3.0% and it closed lower on 61.7%. Split by era, that is −1.3% and 45.3% for the 53 such days before 2020, against −4.3% and 74.6% for the 67 since. Whatever independence bitcoin had on bad equity days belongs mostly to the earlier half of our data.

Stretch the horizon from days to a month and the same thing shows up. Take the ten worst runs of 21 trading days in a row for the index — about a calendar month of trading, picked so that the same slump is never counted twice — and bitcoin was down in nine of the ten, falling 15.1% on average against the index’s 12.7%, while gold finished higher in six. Ten episodes is a small sample and several sit inside the same two years, so read it as a direction, not a rate.

None of these three counts is large and March 2020 appears in all of them — but asked three different ways, they answer the same way.

Claim two: “digital gold”

If the phrase means “behaves like gold in the portfolio”, our data does not support it. The full-sample correlation between the two is 0.093; on a trailing one-year basis the median reading is 0.04, and only 7.2% of all 3,505 windows exceed 0.3 — the highest, 0.41, came on 5 November 2020, and the latest reading is 0.22. The two assets have moved together roughly as often as any two unrelated things do.

Nor do they behave alike under stress: across the twenty worst equity days counted above, gold’s average move was −0.5% and its deepest −5.0%, against bitcoin’s −5.9% and −39.5%.

Where the phrase does hold is in the argument that produces it — a supply schedule nobody can change, no issuer, no cash flow to discount. Those similarities are real and entirely qualitative, and they are also why both assets are hard to value, which is a shared weakness as much as a shared strength. And note what gold’s own fifteen years look like in the table above: 5.4% a year, a 44.6% fall and nearly nine years to recover it. “Digital gold” is not automatically a compliment.

Claim three: “inflation hedge” — we cannot test this

We hold no consumer price series. Every number on this site, including every one above, is nominal: dollars of the day, never adjusted. Testing an inflation hedge honestly needs a price index to deflate by, and we do not publish data we cannot verify from our own sources. So we will not tell you whether bitcoin is an inflation hedge — on our data the claim is untestable.

That gap deserves a minute rather than an apology, because it applies well beyond this site. The claim as usually stated has no measurable content — it names no horizon, no index, no country. And even with a perfect CPI series, fifteen years holds roughly one inflation episode: a relationship confirmed by a single episode is a story, not evidence, whichever way it comes out.

What we can say is modest: over these fifteen years all three assets grew in nominal terms — 78.7%, 15.3% and 5.4% a year — so any inflation rate below the smallest of those leaves all three ahead in real terms. That is far weaker than “hedge”, and it is the strongest version our data supports. So one of the three claims this article set out to test leaves the table unanswered rather than answered — and the fact that a third of the standard case for bitcoin is the part hardest to check against any price history is itself worth knowing. A claim you cannot check is not a claim in your favour; it is one to hold loosely.

Everything above is about bitcoin alone. Almost nobody holds it that way

An 82% volatility is terrifying for an asset that is all of your money and mild for one that is a twentieth of it. So here is the same fifteen years measured on a pair instead of a single asset: 95% in the index, 5% in bitcoin, kept at those weights.

Held at 95% index / 5% bitcoin	Index alone	With the 5% sleeve
From 18 Aug 2011 — annualised return	15.3%	19.7%
From 18 Aug 2011 — volatility	17.0%	17.3%
From 2 Jan 2018 — annualised return	14.2%	15.4%
From 2 Jan 2018 — volatility	19.3%	19.6%
From 4 Jan 2021 — annualised return	14.7%	15.4%
From 4 Jan 2021 — volatility	16.6%	17.1%

The sleeve helped in every window we measured, including the two where bitcoin on its own failed the risk test: +4.4 points of annual return for +0.3 points of volatility since 2011, and a smaller but still positive +1.3 for +0.2 since 2018. That is the strongest thing this article can say for owning some bitcoin. It is also arithmetic rather than a promise — at 5% the sleeve contributes so little to the portfolio’s swings that even a mediocre return per unit of risk still adds, and the same arithmetic runs the other way as the weight grows. Three overlapping windows of one history are not a distribution of futures.

What a long-term holder can take from this

• Treat the headline multiple as a fact about 2011. The 5,864x is inseparable from a $10.90 entry. The honest version — how often bitcoin beat the benchmark over every possible one-, three- and five-year window — is what the comparison tool computes daily.
• Size the position against the risk, not the return. Volatility near 80% a year, a −84.9% record fall and three-quarters of days spent 20% or more below the peak are the terms of the contract. Our volatility history tracks how that number has drifted; the holding-period statistics show how the odds changed with the length of the wait.
• Do not budget for bitcoin to cushion an equity fall. On the twenty worst days for the index it was down on ten and down harder than the index on eight; since 2020 it has fallen on three-quarters of the days the index dropped 2% or more. Whatever it added to a portfolio, it added through return, not through protection on bad days — hold it for the first, not the second.
• Prefer a schedule to a timing decision. The entry date dominates the outcome, and it is exactly the variable nobody controls in advance. Spreading purchases is the standard answer to a dominant unknown; what it did and did not fix is in our DCA guide.

None of this is advice, and none of it is a forecast; it is a description of one window of history that will look different once the next cycle is added to it. What a long-term holder actually owns next to a brokerage account is, on this evidence, neither digital gold nor a hedge: it is a very high-return, very high-volatility asset whose independence from equities has been shrinking, and which — precisely because it is so unlike the rest of the portfolio — has improved that portfolio at small weights in every window we measured. Both halves of that sentence come from the same table. Which one matters more depends on how much of it you own.

Frequently asked questions