Position sizing without going broke

You can be right about a market and still go broke on it. Edge tells you whether to bet; it says nothing about how much. Bet too little and a genuine advantage takes years to show up. Bet too much and a normal losing streak — the kind that is guaranteed to happen — takes you out before the advantage ever pays.

The Kelly criterion, in one line

Kelly gives the stake that maximises the long-run growth rate of a bankroll. For a binary market at a decimal price, the fraction of bankroll to stake is:

f = (bp − q) / b

where p is your probability, q is 1 − p, and b is the net odds received on a win (decimal odds minus 1). On a prediction market priced in cents, buying at price *c* means b = (1 − c) / c.

A worked example

You make something 61%. It is trading at 55c. Buying at 55c pays 45c on a 55c stake, so b = 0.45 / 0.55 = 0.818.

• bp = 0.818 × 0.61 = 0.499
• q = 0.39
• f = (0.499 − 0.39) / 0.818 = 13.3% of bankroll

Thirteen percent of everything you have, on one market. That should feel like too much, because it is.

Why full Kelly is the wrong answer

Kelly is optimal *given that your probability is exactly right*. It is not. Yours is an estimate, and Kelly is brutally sensitive to overestimating your edge — the penalty for betting too big is much worse than for betting too small.

That last row is the one to sit with. Betting twice the Kelly stake on a genuine edge has a long-run growth rate of zero. You can be right about every single market and still end up exactly nowhere, purely through sizing.

A sizing policy you can actually follow

1. 1Compute Kelly on your net edge, after vig, spread and fees — not the gross gap.
2. 2Take a quarter to a half of it. Pick one and keep it. Switching multipliers based on how you feel about a bet undoes the whole point.
3. 3Cap any single position at something you would be relaxed about losing — 2–5% of bankroll is a common ceiling regardless of what Kelly says.
4. 4Cap correlated exposure. Five markets that all resolve on the same election are one bet wearing five hats. Size the theme, not the ticket.
5. 5Never size up to recover. Kelly is a function of edge and price. It has no term for how your month is going.

When Kelly says zero

If your edge is negative, the formula returns a negative number. That is not an instruction to bet the other side — the other side has its own price and its own cost, and it is usually also negative once you have paid the spread twice. It means there is no bet here. Most of the time, on most markets, that is the correct answer.