Fractional-Kelly Position Sizing, Why Quarter-Kelly Survives
Pull-quote: “Full Kelly is optimal only if the edge estimate is exactly right. Quarter Kelly is what optimal looks like once you admit that it is not.”
Why this matters
The Kelly criterion answers a precise question: given a known edge and known payoff odds, what fraction of capital maximizes long-run geometric growth. The mathematics is clean, the logic goes back to Kelly’s 1956 paper and Thorp’s casino work, and almost no systematic desk should run it at full strength. The reason is not timidity. The inputs are estimates, and the criterion punishes overestimation far more severely than underestimation. Position sizing is therefore not a forecasting problem. It is an engineering problem: size positions so that being wrong about your own edge, which you will be, does not end the process.
The growth curve is not symmetric
Write expected growth as a function of the fraction of Kelly you bet. The curve rises to a peak at full Kelly, falls past it, and reaches zero at twice Kelly. Beyond that, growth is negative even though the edge is real.
growth
▲ peak: full Kelly
│ ●●●●●
│ ●●● ●●●
│ ●● ●●
│ ● ●
│ ● ● zero growth
│ ● ● at 2x Kelly
│ ● ●
└──┬─────┬───────┬──────●────► fraction bet
0.25x 0.5x 1.0x 2.0x
quarter half full ruinousThe following are properties of the criterion’s own arithmetic, not claims about any strategy:
| Sizing | Share of maximum growth | Variance of outcomes | If the edge was overestimated 2x |
|---|---|---|---|
| Full Kelly | 100 percent | Highest | You sit at twice true Kelly: zero expected growth |
| Half Kelly | About 75 percent | Roughly half | You sit at true Kelly: the actual peak |
| Quarter Kelly | About 44 percent | Roughly a quarter | You sit at half of true Kelly: the safe slope |
The table is the whole argument. The right side of the peak is a cliff and the left side is a slope. An estimation error that doubles the believed edge moves a full-Kelly bettor to the zero-growth point. The same error moves a quarter-Kelly bettor to half of true Kelly, which remains a comfortable place to operate. Under the standard continuous approximation, a full-Kelly bettor also carries roughly an even chance of halving the bankroll at some point along the way. That is not a tail event. It is the expected texture of the ride.
Edges are estimates, and estimates decay
The case for fractional sizing strengthens once you account for how edges behave in production. They are measured with error on finite samples. They decay as markets adapt. They correlate across strategies in stressed regimes, exactly when sizing matters most. Each of those facts moves the prudent fraction down. None of them moves it up. Quarter Kelly is not a magic constant. It is a defensible point on the trade between growth and survival once estimation error, decay, and correlation are admitted into the model rather than assumed away.
Sizing as a property of the system, not a habit
In production, fractional-Kelly sizing is a rule inside a deterministic risk engine: the same portfolio state and the same instrument data produce the same size, every time. Sizing is liquidity-weighted, so the position the formula wants is capped by what the venue can actually absorb. Hard daily stops sit underneath as the backstop: sizing shapes the distribution of outcomes, and the stop caps the worst day. Every cap and every breach lands in an append-only audit log, so sizing decisions are reviewed later against what was known at the time, not against how the trade turned out.
Closing
Kelly’s insight survives: size positions to the edge, not to the appetite. What does not survive contact with production is the assumption that the edge is known. Quarter Kelly is the version of the criterion that admits uncertainty and stays in business. On a systematic desk it belongs in code, inside a deterministic risk engine, behind hard stops, in front of an audit log. Survival first is not a slogan. It is a sizing policy.