In 1956, a Bell Labs physicist derived the Kelly criterion — the formula
for how much to stake on a favorable bet. Applied to bitcoin, its answer is startling —
and the honest lesson is not the number, but the shape of the curve around it.
Every page on this site about bitcoin's properties eventually meets the same practical question, and it is the one question the literature spends the least honest effort on: not whether bitcoin belongs in a portfolio, but how much. There is, as it happens, a famous mathematical answer to exactly this question — and it did not come from finance.
In 1956, John Kelly, a physicist at Bell Labs, was studying how fast information could travel down a noisy telephone line. He noticed that his problem had the same structure as a gambler's: given a genuine edge and uncertain outcomes, what fraction of your capital should you stake to grow it as fast as possible without ever going broke? The answer — bet in proportion to your edge, discounted by your uncertainty — became known as the Kelly criterion?A rule for position sizing that maximizes the long-run compound growth rate of wealth. For an asset held against a safe alternative, the optimal fraction is the asset's expected excess return divided by the square of its volatility.. Claude Shannon, the father of information theory and Kelly's colleague, took it seriously enough to use it. So did Edward Thorp, who applied it first to blackjack, in the card-counting system that made casinos change their rules, and then to markets, at Princeton Newport Partners — one of the first quantitative hedge funds, which compounded at roughly 20% a year for nineteen years. Bill Gross, who ran the world's largest bond fund, learned position sizing from Thorp's blackjack book. And in bitcoin specifically, no one has done more to popularize the criterion than Fred Krueger, whose Bitcoin One Million (with Ben Sigman) gives the formula a chapter of its own and made it part of how bitcoiners talk about allocation. The formula's pedigree is not in question.
What the formula maximizes is the long-run growth rate of wealth — not the expected value of any single bet. The distinction matters because wealth compounds: a 50% loss requires a 100% gain to repair, so a strategy that looks attractive on average can still grind a single investor to nothing over time. Kelly sizing is the mathematical answer to living in one timeline?The ergodicity point: an "expected value" averages across many imagined parallel outcomes, but a person experiences one sequence of outcomes in order. For compounding wealth, the time-average growth of a single path is what matters — and maximizing it is exactly what the Kelly criterion does. rather than averaging across many.
This page applies that formula to bitcoin, honestly. The raw answer will be startling. The reasons to discount it are just as instructive. Neither is a recommendation; both are worth seeing plainly.
For an asset held against a safe alternative, the Kelly fraction has a clean closed form: the asset's expected excess return?The asset's expected annual return (its "drift," in arithmetic terms) minus the risk-free rate — what a safe alternative such as a Treasury bill pays. The formula sizes the bet on the edge over doing nothing, not on the raw return. divided by the square of its volatility?The standard deviation of the asset's annual returns — how widely outcomes swing around their average. Squaring it gives the variance, the penalty term in the formula. Bitcoin's annualized volatility has historically run between roughly 50% and 80%, several times that of stocks.. One technical care this page takes that most published attempts do not: historical growth is usually quoted as a compound rate, but the formula wants the arithmetic drift, and the two differ by the volatility drag?Volatility makes compound growth lag the simple average of returns: g ≈ μ − ½σ². Plugging a compound growth rate (CAGR) straight into the Kelly formula quietly understates the answer — the most common error in published bitcoin–Kelly calculations. This page converts properly, and shows the conversion in the readout below. — a distinction that quietly decides whether the answer is merely large or calls for leverage.
Run it on bitcoin under three honest input scenarios — computed live on this page from the same Power Law model, coefficients, and price series as the site’s calculators. The two Power Law scenarios are the site’s canonical pair (the same options offered in The Power Law’s sibling calculators), and both embed the trend’s declining growth rate, averaged over the next ten years:
Institutional research lands in the same startling territory when it asks the same question. Fidelity Digital Assets' March 2026 study ran a discrete Kelly calculation on bitcoin's last ten calendar years — seven positive years averaging +288%, three negative averaging −50% — and published the result: a 65% position. Their own footnote concedes they chose the conservative variant of the formula; the generalized version "results in an even higher recommended position size" — it implies leverage. Under their deliberately conservative forward inputs, the same machinery still says 10% — five times the institutional consensus. And note what trend-anchored inputs mean for the most maligned figure in this debate: when the formula's answer exceeds 100%, the unleveraged holder who is "all in" is not being reckless by Kelly's lights — they are under-allocated. The formula is more bullish than the maximalist.
So the startling number is established. Now for the honest part.
The formula's answer is a single point. The argument of this page is the shape of the curve around it. Below is the relationship between how much you allocate and how fast your wealth compounds — with bitcoin's numbers, under assumptions you control. Watch three things: how unevenly the curve falls away on either side of the peak; how violently the peak itself moves when you nudge the assumptions; and how little growth the half- and quarter-Kelly marks give up for how much calamity they avoid.
Three honest reasons, in ascending order of force.
The inputs are unknowable. The formula's answer is exquisitely sensitive to the expected return — the one input nobody has. Chopra and Ziemba's classic study put numbers on it: errors in the return estimate damage outcomes roughly eleven times more than errors in the volatility estimate, and the penalty grows with risk tolerance — the more aggressively you size, the more your mistake costs. And bitcoin's expected return is genuinely contested: a 2026 analysis of the power-law model found that the fitted exponent — the engine of every long-horizon extrapolation — varies by nearly a factor of three depending on a methodological choice as small as where you place the curve's time origin. The fit is real; the precision is not.
The curve is asymmetric. Allocate half the optimal fraction and you keep 75% of the achievable growth. Allocate double and your growth falls all the way back to the risk-free rate — you carry twice the volatility of the optimal portfolio and are paid nothing for it — and beyond double, growth turns negative: more risk, less wealth, forever. Under-betting costs a little; over-betting costs everything. Since the inputs can only be estimated, and since estimation error in one direction is survivable and in the other direction is not, deliberately sitting below the formula is not timidity. It is the rational response to not knowing.
The drawdowns are unsurvivable. At the full Kelly allocation, the mathematics is explicit: a coin-flip's chance — 50% — that your wealth is cut in half at some point along the way, and one chance in three that the halving comes before the first doubling. These are not tail risks; they are the design. The model assumes you hold through them, withdraw nothing, retire never, and feel nothing. Every way in which a person differs from that assumption — a finite horizon, a retirement date, living expenses, a family, loss aversion, the 3 a.m. urge to sell — corrects the allocation in the same direction: downward. For a single investor, a deep enough drawdown is not a statistical outlier to be averaged away. It is an absorbing barrier?A point of no return in a random process. In investing: the loss deep enough that the investor — for financial or psychological reasons — exits and never recovers the position. Once absorbed, the long-run mathematics no longer applies to you. — the point where the game ends, whether or not the math says it should have continued.
The practitioners' answer to all three problems is the same, and it has been standard since Thorp: bet a fraction of the formula?"Half-Kelly" means allocating half of what the formula says; "quarter-Kelly," a quarter. Because the growth curve is flat near its peak and the risk falls fast as you step left, fractional sizing gives up a little growth for a large reduction in drawdown risk.. The arithmetic of the trade is remarkably kind.
At half-Kelly, you retain 75% of the achievable excess growth — while the chance of ever seeing your wealth halved collapses from 50% to 12.5%, and the chance of halving before doubling falls from one-in-three to one-in-nine. At quarter-Kelly you still retain about 44% of the maximum growth, and the halving risk that defines full Kelly's brutality essentially vanishes — well under 1%. The exchange rate, in plain terms: give up a quarter of the growth, shed three-quarters of the catastrophe.
And there is a deeper justification than prudence. The full formula is optimal only for an investor with a mathematically specific — and humanly rare — indifference to swings in wealth. For an investor with ordinary, stronger risk aversion, the formally optimal policy is exactly a fixed fraction of Kelly: the same mathematics, run with an honest utility function, produces the discount directly. Fractional Kelly is not the formula watered down. It is the formula told the truth about its owner.
And the curve answers a question it never draws: does it matter where bitcoin sits on the Power Law trend when you take the position? In this model it does, and in exactly the direction intuition suggests. The drawdown odds in the readout depend on one thing only — the fraction of Kelly you hold, not the asset's raw numbers. Position-on-trend enters through the expected return: enter at a deep discount and the Revert scenario's catch-up growth raises the formula's answer, so any fixed allocation becomes a smaller fraction of Kelly, with materially better drawdown odds. Enter far above trend and the same allocation quietly becomes a larger fraction of a smaller answer. A given weight really is safer at the floor of the channel than at the top — not as folklore, but as the model's own arithmetic, and it is the sizing-language version of what The Power Law's bands express.
One more property worth naming: a Kelly weight is a target, not a transaction — and notice what holding the target means in practice. When bitcoin surges, your slice grows past its weight and the discipline says trim; when bitcoin falls, the slice shrinks and the discipline says add. Operating the Kelly criterion is mechanical sell-high, buy-low — which is precisely the protocol of this page's sister exploration, Disciplined Rebalancing. Kelly answers how much; rebalancing answers how to stay there — and staying there is itself a small, real engine of return, harvesting the volatility that the sizing math treats only as drag. The Power Law's upper band — the spike zone where this site discusses trimming — belongs to that maintenance discipline, not to sizing: a held weight already trims into strength mechanically, and sizing a permanent allocation to a temporary spike is precisely the over-betting the curve punishes.
Hold the fractional discount in one hand and conventional guidance in the other, and the page's finding appears. The same asset, the same arithmetic, descending a ladder of answers:
| The question asked | The answer |
|---|---|
| Full Kelly, revert to Power Law trend (computed live above) | — |
| Full Kelly, stay at current trend multiple (live) | — |
| Fidelity's discrete Kelly, last ten calendar years (Mar 2026) | 65% |
| Half-Kelly, trend-multiple inputs (live) | — |
| Quarter-Kelly, trend-multiple inputs (live) | — |
| Fidelity's discrete Kelly, conservative forward inputs | 10% |
| Maximum-Sharpe optimization, Fidelity's conservative assumptions | 9.4% (and 0% bonds) |
| BlackRock's risk budget — and its model portfolios since 2025 | 1–2% |
| Typical advisor guidance | 1–5% |
The gap between the top and bottom of this ladder is not a disagreement about bitcoin. Every rung can accept the same inputs. It is a disagreement about the question. The bottom rungs ask: how much bitcoin can a portfolio hold before its volatility dominates the risk budget? — a question about smoothness, careers, and committees, with 1–2% as a defensible answer. The top rungs ask: what allocation maximizes how fast wealth compounds? — a question about growth, with answers that start at 10% under hostile assumptions and run past 100% under trend-anchored ones. A single institution's single document carries both ends: Fidelity's March 2026 study answers the Sharpe question with 9.4% and the growth question with 65%, eleven pages apart.
The reader does not have to pick a rung. But it is worth knowing which question each number is answering — because the most commonly cited figures answer the question almost nobody who owns bitcoin is actually asking.
This page has not told you how much bitcoin to own. It has shown you what one well-credentialed body of mathematics says when asked — and why the people who trust that mathematics most are precisely the ones who discount it before acting. Three closing observations, stated descriptively.
First: every honest correction runs in one direction. Unknowable inputs, finite lifespans, real expenses, human nerves — each argues for sizing below the formula. None argues for zero. The same arithmetic that refuses to endorse leverage also refuses, under any input this page can defend, to endorse nothing.
Second: the formula's answer is not shrinking. Because the allocation rises as variance falls, bitcoin's maturation cuts the formula's denominator faster than fading returns cut its numerator across a wide range of assumptions — the math's answer can grow as the asset calms. The institutional ladder has begun to reflect this: the world's largest asset manager moved from publishing 1–2% to allocating it.
Third: whatever fraction a person arrives at, it is a number they must be able to hold — through the drawdowns the curve promises, with the maintenance discipline that keeps a target a target. How much is a question about mathematics. Whether you can keep it is a question about you.
The formula does not say what to do.
It shows the limits of what can be survived.
Every page on this site has been improved by someone pushing on it. Ask a question, flag an error, or suggest what’s missing — it goes straight to the author, never published.