How much to stake: Kelly sizing for prediction markets, simply

Most traders who find a real edge still lose money for one reason: they put too much on a single idea. One bad run and the bankroll is gone before the edge has time to work. The fix is old and simple — size every trade to the edge.

The Kelly rule in one line

On a prediction market you buy a share at price c that pays $1 if the outcome happens. If you think the real probability is p, the Kelly rule says the share of your bankroll to put in is:

f = (p − c) ÷ (1 − c)

No edge (p equal to c) means f is zero — do not trade. A bigger edge or a cheaper price means a bigger stake.

A worked example

The market sells YES at 41¢. You think the real chance is 45%.

f = (0.45 − 0.41) ÷ (1 − 0.41) = 0.04 ÷ 0.59 ≈ 6.8%

Full Kelly says put 6.8% of your bankroll on it. With $2,000 that is about $136.

Why we never use full Kelly

Kelly assumes you know p exactly. You never do. If your 45% is really 42%, full Kelly is already far too big, and the swings are brutal even when you are right — halving your bankroll along the way is normal. Professionals use a fraction:

• Quarter Kelly keeps most of the long-run growth with a fraction of the swings. In the example: about 1.7%, or $34.
• Eighth Kelly for estimates you trust less — for example a model or an AI judgement rather than a liquid market price.
• A hard cap — no single trade above 5% of the bankroll, however good it looks.

Two more limits

• The order book. If the edge only holds for the first $300 of the book, a $500 stake buys the rest at fair value or worse. Stake the smaller of the two.
• Correlated trades. Three signals on the same match are one bet, not three. Size them together.

How DeeplyticAI does it

Where a signal has a probability behind it, the Before you trade block suggests a stake: a quarter of Kelly when the probability comes from sharp bookmakers or another exchange, an eighth when it comes from our AI estimate, never more than 5%. Enter your bankroll once and it shows the amount in dollars — capped by what the order book can actually fill. Signals without a probability, like whale moves, show liquidity only.

Analysis, not financial advice. Only trade money you can afford to lose.