Size kills more edges than bad forecasting. A 70% accurate trader running Kelly at 3x their bankroll underperforms a 75% accurate trader running half-Kelly. The math is not close. If you are trading prediction markets without a sizing model, you are flying on feel.
What the Kelly Criterion says
Kelly gives you the optimal fraction of your bankroll to put on a single trade.
f equals (b times p minus q) divided by b.
f is the fraction to bet. b is the net profit per dollar risked. For a Kalshi or Polymarket binary contract at 64 cents, b equals 0.36 divided by 0.64, or 0.5625. p is your estimated true probability. q is 1 minus p.
Say you estimate 70% true probability on a contract trading at 64 cents. Kelly says: (0.5625 times 0.70 minus 0.30) divided by 0.5625 equals 0.167. Bet 16.7% of bankroll. That is the number before fees, and before any honesty about how good your probability estimate actually is.
Why every serious trader discounts Kelly
Full Kelly maximizes long-run geometric growth. It also produces brutal drawdowns the moment your probability estimates are off, and they are always at least a little off. Your edge is the gap between your estimate and the market's implied probability; measure that gap honestly before sizing anything.
The standard in professional quantitative trading is half-Kelly or quarter-Kelly. Take the raw fraction and cut it in half before you place the order.
- Full Kelly: theoretically optimal growth, maximum variance. One bad estimate wrecks you.
- Half-Kelly: where most practitioners start. Cuts drawdowns significantly with modest sacrifice to long-run growth.
- Quarter-Kelly: standard for thin books, uncertain estimates, or strategies you have not run long enough to trust.
A probability estimate of 70% that is actually 62% destroys a full-Kelly position. The same error on quarter-Kelly leaves you trading tomorrow. Survival is part of the strategy.
Fees do not negotiate with your model
Taking liquidity on Kalshi costs roughly 1% to 2% per side. That is a 2% to 4% round-trip. On a contract at 64 cents where your gross edge is 6 percentage points, you hand 3 to 4 of those points to the venue before you exit the trade.
The fee-adjusted calculation on that same trade: (0.70 times $0.36) minus (0.30 times $0.64) minus $0.03 round-trip equals $0.03 net per dollar wagered.
Still positive. But once the $0.03 fee is baked into the entry cost, the Kelly fraction drops from 16.7% to around 9%. Fees nearly cut the position size in half. The position your model calculates before fees and the position you should actually take are different, and the difference is fees.
Maker fees, where the venue offers them, cut this friction. Know the fee structure of every contract before you size. Fees are not optional math.
The bet the math says to skip
A sizing model also tells you when not to trade.
If fees consume your gross edge, the net Kelly fraction is zero or negative. Zero means do not bet. That is not the model failing. That is the model working.
Prediction markets list hundreds of open contracts. The discipline is not finding one you like. It is finding one where the math is positive after fees, the liquidity absorbs your size without moving the price, and your probability estimate reflects an edge you actually have.
A contract at 64 cents that every other trader also reads as 64 cents has no edge in it. Pass it. The model returns zero and so should you.
Sizing across multiple positions
A single-position Kelly fraction assumes your capital is deployed in one bet at a time. Prediction markets do not work that way. On an active news cycle, a trader might hold positions on an inflation print, an election primary, and a central bank decision simultaneously.
Positions on uncorrelated events can each carry their standalone Kelly fraction. The errors in your probability estimates are independent, so the aggregate risk stays bounded. A bad estimate on one position does not compound with the others.
Correlated positions are different. Two contracts that both resolve on the same candidate's performance, or two macro-event contracts that both depend on the same economic print, move together. Running full Kelly on each independently overstates your aggregate edge. If your core assumption is wrong, both positions lose. Cut the individual fractions in proportion to the correlation.
Kairos shows your open positions across Kalshi, Polymarket, and Predict.fun in one view. Spotting correlation risk is a visual problem before it becomes a math problem. Positions you would not size aggressively if you saw them on the same screen often end up at full Kelly when they live in different tabs.
Venue price sits upstream of every other number
The same contract trades at different prices across Kalshi, Polymarket, and Predict.fun. A 64-cent contract on one venue might sit at 61 on another. That 3-cent gap shifts your effective edge by 3 full percentage points.
On a 6-point gross edge, trading at 61 cents instead of 64 cents determines whether the trade is positive-EV or negative-EV. Traders running prediction market arbitrage already know that venue price selection is where the edge lives. The sizing model is downstream of the price. Get the wrong price and every number that follows is built on a wrong input.
Kairos aggregates Kalshi, Polymarket, and Predict.fun into one book. Global best bid and best ask, across every venue, in a single view. The right price is not the first price you see. It is the best price in the market, and Kairos shows you both at once. Sizing discipline shows fastest under pressure — the Kairos Cup is where traders test it head-to-head.
The power of an institutional trading desk, on your laptop. Size correctly. Execute at the best price in the market. Kairos.