Guide · Part II — The mental tools
Kelly and fractional sizing
A genuine, positive edge plus an over-sized bet equals eventual ruin. That single sentence is the reason position sizing is a separate discipline from asset selection, and the Kelly criterion is the mathematics that proves it.
The problem
Every allocator eventually faces the question the previous two chapters cannot answer: given an attractive opportunity, how much? Markowitz says how to mix; the compounding arithmetic says why losses hurt disproportionately; neither names a number. Intuition fills the gap badly — conviction inflates size, and humans systematically overestimate their edge.
The stakes are asymmetric in a specific way. Bet too small and the cost is mild: slower growth. Bet too big and the cost is terminal: one ordinary losing streak takes a bite that compounding can never grow back. Sizing is therefore not a detail of a strategy; it is the difference between a strategy surviving and not existing.
The insight
John Kelly, working at Bell Labs in 1956, showed that for a repeated favourable bet there is a single fraction of capital that maximises long-run compound growth. The formula is friendlier than its reputation: bet the win-probability minus the loss-probability divided by the payout ratio.
A worked example makes it concrete. Take a favourable coin: it wins 60% of the time, and a win pays twice the amount risked. Kelly says bet 0.6 minus 0.4-divided-by-2, which is 0.4 — risk 40% of capital each round. That 40% is “full Kelly.” Half-Kelly is 20%, quarter-Kelly is 10%, and all-in is 100%. Hold those four numbers.
Now run twenty rounds of this genuinely favourable game, starting from any stake. All-in survives only if all twenty flips win: 0.6 to the twentieth power, roughly four chances in a hundred thousand. Near-certain ruin, despite a real edge — because a single loss at 100% sizing is fatal, and the edge only pays those still in the game. Full Kelly grows fastest on paper but with gut-churning swings; a couple of early losses can halve the stake. Half-Kelly compounds into multiples of the start in a representative run, with far shallower drawdowns. Quarter-Kelly is slower and calmest, almost never seriously dented.
The practitioner standard is half or quarter Kelly, and the reason is not timidity but epistemology. Full Kelly is optimal only when the edge is known exactly, and in markets the edge is always an estimate — usually an optimistic one. Betting full Kelly on an overestimated edge is betting over Kelly on the true one, and over-Kelly sizing does not merely add volatility; it turns a winning game into a losing one. The growth-versus-size curve is asymmetric: to the left of the optimum it slopes gently, to the right it falls off a cliff. Fractional sizing is the margin of safety for not knowing the true edge.
In plain English
Direction and size are two separate decisions, and the second one is where survival lives.
An allocator can be right about the idea and still go broke on the sizing — right-and-ruined is a standard failure mode, not an exotic one.
The honest default size is small, precisely because confidence in the edge is the least reliable input in the whole calculation.
Applied without any formula, Kelly thinking becomes three habits.
- Survivable loss. Size every position so that its total loss would be survivable without changing any life plans.
- Scale to evidence, not excitement. An uncertain edge gets quarter-Kelly treatment, which for most real opportunities means single-digit portfolio percentages.
- Zero is an acceptable outcome. Cap speculative positions at a level where zero is survivable, because for the most uncertain bets, zero is a scenario, not a hypothetical.
The same arithmetic operates at every altitude. Institutions and even nations blow up the same way: a real economic engine, levered too hard, destroyed by an ordinary bad stretch it could not survive at that size. Over-betting a genuine edge is one of the oldest failure patterns in finance precisely because the edge is real — the post-mortem always shows the idea was right.
Where this breaks
Kelly’s assumptions deserve to be stated plainly, because markets violate most of them. The formula assumes known, stable probabilities and payouts; markets offer estimates that drift. It assumes sequential, independent bets; portfolios hold simultaneous, correlated positions, which shrinks the safe aggregate size further. It assumes the capital pool is the whole story; real allocators have external liabilities and time horizons that can force exits at the worst moment.
Kelly also optimises growth, not comfort — even half-Kelly drawdowns exceed what most people can psychologically hold, and a plan abandoned mid-drawdown performs worse than a smaller plan followed. The formula’s real gift is not the number; it is the shape of the curve — the proof that over-sizing is always the expensive side of the error.
The discipline layer later in this curriculum turns this chapter into standing policy: evidence builds, size follows, and no single position is ever allowed to be fatal. The immediate action is simpler. For every current holding, ask one question: if this went to zero, would the plan survive? Any position where the answer is no is over Kelly — whatever the conviction behind it.