How much bitcoin?
Fifteen tested claims on this site say almost nothing tells you when to buy. This page asks the question that survives, how much to hold, and lets you answer it yourself. Move the weight and everything below recomputes from the published monthly series.
Jul 2015 to Sep 2026 · 134 monthly returns · 54% stocks, 36% bonds, 10% bitcoin · rebalanced monthly
At 10% bitcoin in stocks and bonds, the weight added between +1.6 points a year (entered 2022) and +6.5 points (entered 2015). The only difference between those rows is when the money went in.
These are the published figures at a 10% weight in 60% stocks and 40% bonds. The interactive version, which lets you change the weight, the base, the start year and the rebalancing rule, needs JavaScript.
Growth of one unit, log scale. The chosen mix against the same portfolio with no bitcoin.
What the weight bought, and what it cost
Each dot is an entry year. Higher means the bitcoin weight added more annual return; further right means it made the worst fall deeper. 7 of the 9 stack in one column: everyone who entered from 2015 to 2021 went through the same fall, which bottomed in October 2022, so they paid the same 6.2 extra points. What they gained still ranged from +1.8 to +6.5. When you started changed the reward far more than the risk.
| Entered | No bitcoin | With bitcoin | Added | Extra worst fall |
|---|---|---|---|---|
| 2015 | 8.9% | 15.4% | +6.5 | -6.2 pts |
| 2016 | 9.7% | 16.0% | +6.4 | -6.2 pts |
| 2017 | 9.4% | 15.6% | +6.1 | -6.2 pts |
| 2018 | 8.8% | 11.6% | +2.8 | -6.2 pts |
| 2019 | 10.3% | 15.2% | +4.8 | -6.2 pts |
| 2020 | 8.8% | 12.9% | +4.1 | -6.2 pts |
| 2021 | 7.8% | 9.6% | +1.8 | -6.2 pts |
| 2022 | 7.1% | 8.7% | +1.6 | -4.1 pts |
| 2023 | 13.0% | 16.7% | +3.7 | -0.3 pts |
Only entry years with at least three years of returns are shown. A shorter stretch, annualised, looks precise and means very little.
Every window, not just calendar years
Entry years are 9 starting points. This takes every month as a starting point. Each line shows what the weight added a year over the window that starts in that month. Over 99 three-year windows, 10% bitcoin never cost return; the least it added was +0.7 points, from October 2021. This sample has no three-year stretch in which bitcoin lagged badly enough to drag the whole portfolio down. A future one could.
| Held for | Windows | Worst added | Median added | Best added | Windows it lost | Worst extra fall |
|---|---|---|---|---|---|---|
| 3 years | 99 | +0.7 Oct 2021 | +5.6 | +15.0 Sep 2015 | 0 of 99 | -7.7 pts |
| 5 years | 75 | +0.7 Mar 2021 | +5.8 | +13.2 Mar 2016 | 0 of 75 | -6.6 pts |
| 10 years | 15 | +6.0 Jul 2016 | +6.7 | +7.9 Sep 2015 | 0 of 15 | -6.2 pts |
Windows overlap, so they are not independent: the ten-year row is fifteen windows sharing most of their months, which is close to one observation.
Does the base change the answer?
The same 10% weight, with stocks at 60% of the rest. Next to bonds it added +6.5 points a year; next to gold, +6.3. The base moves the level of return a great deal and what bitcoin added hardly at all. Leaving out dividends, as this page first did, makes the whole portfolio look about 1.1 points a year worse and changes what bitcoin added by only 0.06.
| Base | Return, none | Return, with | Added | Worst fall, with | Extra worst fall | Sharpe, with |
|---|---|---|---|---|---|---|
| 60% stocks, 40% bonds | 8.9% | 15.4% | +6.5 | -24.8% | -6.2 pts | 1.18 from 0.89 |
| 60% stocks, 40% gold | 14.0% | 20.3% | +6.3 | -22.1% | -7.2 pts | 1.44 from 1.25 |
| 60% stocks without dividends, 40% gold (the original base) | 12.9% | 19.2% | +6.3 | -22.7% | -7.1 pts | 1.37 from 1.14 |
How much monthly averages hide
Measured on daily closes from October 2016, every worst fall is deeper: by 2.5 to 3.6 points for the portfolio, and by much more for stocks alone, whose 2020 crash lasted about a month, too short for a monthly average to catch. The comparison between weights holds: 10% bitcoin deepened the worst fall by 5.6 points on daily closes and 6.1 on monthly averages. Read every worst fall on this page as a few points too shallow.
| Holding | Monthly averages | Daily closes | Understated by |
|---|---|---|---|
| 60/40 stocks and bonds, no bitcoin | -19.2% | -22.2% | 3.0 pts |
| with 10% bitcoin | -25.3% | -27.8% | 2.5 pts |
| with 20% bitcoin | -31.3% | -34.9% | 3.6 pts |
| S&P 500 alone | -20.2% | -33.9% | 13.7 pts |
| Bitcoin alone | -76.7% | -83.2% | 6.5 pts |
Stocks and bonds only: there is no free daily gold series, so the gold bases cannot be checked this way. The daily stock leg is the S&P 500 price, from FRED, which starts in late 2016.
Is it still a diversifier?
The 36-month correlation of bitcoin to stocks was 0.28 in 2018 and is 0.38 now; to bonds it is -0.21, and to gold -0.03. The stock figure is the one to watch: an asset that moves with stocks does less diversifying work than the long-run figures on this page suggest.
A weight that moves with the assets it is meant to diversify does less work than the long-run figures suggest. This is the number most likely to change the answer above, and it is not stable.
The published ladder
135 monthly prices, 134 monthly returns, July 2015 to September 2026. Every series is a monthly average. Cash, for Sharpe and Sortino, averaged 2.2% a year.
| Bitcoin | Return a year | Volatility | Sharpe | Sortino | Worst fall | Longest underwater |
|---|---|---|---|---|---|---|
| none | 8.9% | 7.5% | 0.89 | 1.30 | -18.6% | 25 mo |
| 1% | 9.6% | 7.7% | 0.95 | 1.40 | -19.1% | 25 mo |
| 2% | 10.2% | 7.9% | 1.01 | 1.49 | -19.7% | 25 mo |
| 5% | 12.2% | 8.8% | 1.11 | 1.74 | -21.5% | 26 mo |
| 10% | 15.4% | 10.9% | 1.18 | 2.03 | -24.8% | 26 mo |
| 20% | 21.8% | 16.3% | 1.17 | 2.31 | -31.0% | 27 mo |
- No optimiser, no recommended weight. Optimising on 134 monthly returns produces answers that swing wildly on small changes in assumed return.
- Before and after costs. The ladder is gross. The tool's "after costs" figure charges the bitcoin sleeve a spread on every month's rebalancing trade and a flat annual custody fee; the other legs are assumed to carry the costs they already do. Both default to 0.5%, ordinary for a spot ETF plus a broker, and both go to 3%.
- Sharpe and Sortino. Both are shown because they disagree. Sortino penalises only falls, which flatters an asset whose big moves are mostly upward. Both use the actual cash rate each month, not a flat assumption.
- Longest underwater. Consecutive months below a previous peak. This is what makes people sell.
- Checked independently. Every figure in this table, the entry-year table and the rolling windows is recomputed by a second program that shares no code with the pipeline, verify_portfolio.py. Anyone can run it against the published file.
What this does not show
Bitcoin went from about $300 to about $80,000 inside this sample (monthly averages). Every weight looks good in aggregate for that reason alone. The entry-year and rolling tables exist because the aggregate hides how unevenly that arrived.
Every series is a monthly average, not a month-end close. It has to be: the stock and gold series exist only as monthly averages. Averaging smooths, so volatility and the worst fall are understated; the daily check above shows by how much where it can be measured.
Stocks are the S&P 500 with dividends reinvested, from Robert Shiller’s series. His dividend figures stop in June 2023, so months from July 2023 carry an estimate, grown at the dividend’s own ten-year rate of 7.5% a year. The “without dividends” base is the page’s original one, kept so earlier links and figures can still be checked.
Bonds are a 10-year US Treasury held at constant maturity, priced each month from the published yield. That is the standard way to build a bond index from yields; it is not a fund you could buy, and it carries no fees.
Cash, for Sharpe and Sortino, is the effective federal funds rate each month.
This is not advice, and cannot be
These are historical distributions from a short, exceptional sample: 134 monthly returns, about one and a half bitcoin cycles. What happened is not an estimate of what will.
They know nothing about your horizon, your tax position, your other assets, or what you would actually do in a thirty per cent fall, which is the variable that decides most outcomes.
Recomputed daily by fetch/portfolio.py and published as portfolio.json. The tool above recomputes in your browser from the same monthly series, and a test checks it against the ladder every build. Method on the methods page; the claims that motivated this page are on the scorecard.