Photo by GuerrillaBuzz on Unsplash
The short version
- Full Kelly maximizes long-run compound growth in theory, but it assumes you know your edge exactly — a condition that essentially never holds with a volatile ETF sleeve.
- Half-Kelly captures roughly three-quarters of the growth rate at about half the volatility, which is why practitioners size below the theoretical optimum rather than at it.
- Bottom line: fractional Kelly is less a formula than a discipline for keeping a high-variance satellite small enough that estimation error can't quietly ruin the core.
Most sizing debates about a speculative sleeve — a leveraged fund, a thematic AI bet, a single-factor tilt — never get past a round number. Ten percent feels aggressive, five percent feels safe, and the choice is made by intuition rather than by any relationship between the position's edge and its variance. Fractional Kelly is the tool that replaces the round number with a computation, and then deliberately backs off from it. The central question of this article is narrow and practical: given an uncertain edge and real volatility, how large should a satellite position actually be, and why is the mathematically optimal answer almost always too large?
What the Kelly criterion actually says
The Kelly criterion, introduced by John Kelly in 1956 and later championed by Ed Thorp, answers one question: what fraction of capital, bet repeatedly, maximizes the long-run growth rate of wealth? For a simple bet with win probability p, loss probability q = 1−p, and even-money payoff, the optimal fraction is f* = p − q. For continuously compounded returns — the relevant case for an ETF — the analogue is cleaner and more useful:
f* = μ / σ²
where μ is the expected excess return over the risk-free rate and σ² is the variance of returns. The formula has an intuitive shape. Edge in the numerator pushes you to bet more; variance in the denominator, squared, pushes you to bet dramatically less. Doubling your assumed edge doubles the position. Doubling the volatility cuts it to a quarter. Variance dominates, which is the first uncomfortable lesson for anyone sizing a high-volatility sleeve by feel — feel tends to weight expected return and underweight variance.
A worked example makes the scale concrete. Suppose a satellite holding carries an expected excess return of 8% annually with 40% annualized volatility — not unreasonable for a leveraged or narrow thematic ETF. Then f* = 0.08 / 0.40² = 0.08 / 0.16 = 0.50. Full Kelly says put half your capital there. Almost no disciplined investor would, and the reason they're right to refuse is the entire point of the fractional approach.
Why full Kelly is the wrong target
Full Kelly is optimal only under an assumption that never survives contact with markets: that you know μ and σ exactly. You don't. You estimate them from a finite, single-regime history, and the estimate of expected return in particular is extraordinarily noisy — far noisier than the estimate of volatility. This matters because the growth-rate function around the Kelly optimum is asymmetric. Betting below f* costs you a little growth. Betting above f* costs you a lot, and beyond 2× Kelly the long-run growth rate turns negative even though every individual bet still has positive expectation. You can be right about direction and still compound toward zero purely through oversizing.
Overestimate your edge by a factor of two — a routine error when the sample is short — and full Kelly on the true parameters becomes double Kelly, sitting exactly on the line where growth vanishes. That is the estimation-error trap. The correct response is not better forecasting; forecasts of expected return don't improve enough to matter. The correct response is to size as if your edge is smaller than you think, which is precisely what a fractional multiplier does.
Full Kelly is a ceiling computed from numbers you don't actually know; fractional Kelly is the acknowledgment that the numbers are guesses and the ceiling should be treated as a place you never go.
The fractional multiplier and what it buys
Fractional Kelly scales the optimal bet by a constant k between 0 and 1: you hold k · f*. The most-cited choice is half-Kelly (k = 0.5), and the reason it's cited is a clean result from the growth-rate mathematics. Because the growth rate is approximately quadratic near the optimum, betting half of f* retains about three-quarters of the maximum growth rate while cutting the volatility of the wealth path by roughly half. You give up a quarter of the theoretical growth to remove half the swing — and, more importantly, to buy a wide margin of safety against having overestimated your edge in the first place.
| Multiplier | Position (from f*=0.50 example) | ≈ Share of max growth | ≈ Share of full volatility |
|---|---|---|---|
| Full Kelly (1.0×) | 50% of capital | 100% | 100% |
| Half Kelly (0.5×) | 25% of capital | ~75% | ~50% |
| Quarter Kelly (0.25×) | 12.5% of capital | ~44% | ~25% |
Notice the quarter-Kelly row. It corresponds to holding 12.5% of capital in the example — still, to my mind, a large allocation for a single volatile satellite, which tells you how conservative real-world sizing needs to be relative to the raw formula. In practice many disciplined allocators run quarter-Kelly or lower on a speculative sleeve and treat even that as a cap, not a target.
Volatility drag is the mechanism, not a footnote
The non-obvious point that fractional Kelly encodes is why oversizing destroys growth even with positive expected returns: volatility drag. Compound growth depends on the arithmetic mean return minus roughly half the variance. For a 40%-volatility holding, that variance penalty is on the order of 8% per year (0.5 × 0.40²), before you've adjusted for any leverage-induced path dependency. Size that holding too large and the variance term, which scales with the square of your position, overwhelms the linear return term. This is the same arithmetic that governs why levered products compound below their stated multiple over choppy paths — a mechanism I've walked through in the honest math of leveraged ETFs and in the VOO/QQQM/TQQQ comparison. Kelly sizing and volatility decay are two views of one equation: both say that variance is a cost that compounds, and both say that the correct defense is a smaller position rather than a better forecast.
This also reframes what "risk" means in the sizing decision. The danger in a satellite sleeve is not the day-to-day quotation swing but the possibility that oversizing turns a positive-expectation holding into a negative-growth one — closer to permanent impairment than to temporary volatility, a distinction worth keeping explicit, as in the discussion of volatility versus permanent loss.
Applying it without pretending to precision
A workable procedure looks like this. Estimate the satellite's excess return and volatility from the longest history available, and be honest that the return estimate is the weak link. Compute f* = μ/σ². Apply a fractional multiplier no larger than one-half, and lower — quarter-Kelly or below — when the edge rests on a short live track record, a single market regime, or a backtest vulnerable to data-mining. Treat the result as a maximum weight, not a mandate to reach it. Then bound the sleeve independently at the portfolio level so that a total loss of the satellite is survivable for the long-horizon core, and rebalance the position back toward its target rather than letting a winning satellite grow into a concentration you never chose.
The current regime argues for restraint on the edge side of the fraction. With the 10-year Treasury near 4.58% and the effective fed funds rate at 3.63% (FRED, asof 2026-07-14 and 2026-06-01), the risk-free hurdle a satellite must clear is meaningfully higher than in the prior decade, which shrinks μ — the numerator — for any given gross return assumption. A subdued VIX of 16.5 (FRED, asof 2026-07-14) understates the tail risk embedded in a levered or thematic holding; realized volatility in a stress event can be several multiples of the calm-regime figure, and it is stress-regime variance, not today's, that determines whether your sizing survives. When the hedging question becomes acute, the sleeve-level guardrails matter as much as the Kelly fraction itself, a theme developed in building a hedge before the volatility arrives.
FAQ
Is the Kelly criterion appropriate for a buy-and-hold ETF investor?
Kelly optimizes long-run compound growth, which aligns with a long-horizon objective. The caveat is that it assumes repeated, independent bets and known parameters. For an ETF sleeve, treat it as a sizing discipline that caps the position, not as a signal to bet the full computed fraction.
Why half-Kelly specifically?
Because the growth-rate function is approximately quadratic near its peak, halving the bet keeps roughly three-quarters of the growth while cutting volatility of the wealth path by about half. It's a favorable trade when your inputs are uncertain — which they always are.
What happens if I bet more than full Kelly?
Growth falls on both sides of the optimum, but the penalty above it is severe. Beyond about 2× Kelly the long-run growth rate turns negative despite each bet having positive expected value. Overestimating your edge is the common way investors unknowingly cross that line.
How do I estimate the edge for a volatile ETF?
You estimate μ and σ from history, accepting that the expected-return estimate is very noisy and regime-dependent. Because the estimate is unreliable, the honest response is to shrink the assumed edge and apply a smaller fractional multiplier rather than to trust a point forecast.
Does fractional Kelly replace a portfolio-level position limit?
No. Use both. Kelly sizes the position relative to its own edge and variance; a separate portfolio cap ensures that a total loss of the satellite remains survivable for the core. They answer different questions and should be applied together.
Key takeaways
- The continuous Kelly fraction is f* = μ/σ²; variance in the denominator means volatility dominates the sizing decision far more than expected return.
- Full Kelly is optimal only with known parameters, so it is the wrong target in practice — estimation error in expected return makes it dangerous.
- Half-Kelly retains roughly 75% of the growth at about 50% of the volatility; quarter-Kelly or lower is defensible for a satellite resting on a short or single-regime track record.
- Volatility drag is the mechanism behind the whole framework: oversizing turns positive-expectation holdings into negative-growth ones.
- Pair the Kelly fraction with an independent portfolio-level cap and disciplined rebalancing so a satellite's total loss never threatens the long-horizon core.
Methodology. Kelly and fractional-Kelly relationships follow the standard growth-rate framework (Kelly 1956; Thorp). The growth-versus-volatility shares cited for half- and quarter-Kelly use the quadratic approximation of the growth-rate function near its optimum and are illustrative rather than exact. Macro figures are from FRED: 10-year Treasury 4.58% and VIX 16.5 (asof 2026-07-14), effective fed funds 3.63% and CPI 3.7% year-over-year (asof 2026-06-01), pulled 2026-07-16. No individual ETF price data was used in this framework piece.
This article is for educational purposes and does not constitute personalized financial advice. See our full Disclaimer.