Kelly Criterion for Polymarket: How to Size Bets Without Blowing Up
Having an edge means nothing if you bet too much and go bust. Kelly Criterion is the mathematically optimal way to size positions on prediction markets.
Why bet sizing matters more than finding edge
Most traders obsess over finding markets that are mispriced. That's important — but it's only half the equation. A trader with genuine edge who bets too aggressively will eventually go broke. A trader with modest edge who sizes correctly will compound wealth indefinitely.
The Kelly Criterion, developed by John L. Kelly Jr. at Bell Labs in 1956, solves this problem. It tells you the exact fraction of your bankroll to bet to maximise the long-run growth rate of your capital.
The formula
For a binary market (Yes/No), the Kelly fraction is:
Where:
- f = fraction of bankroll to bet
- b = net payout per unit staked (if you buy Yes at 20¢ and it resolves Yes, b = 80/20 = 4.0)
- p = your estimated probability of winning
- q = 1 − p (probability of losing)
Worked example
You find a weather market: "Will Chicago reach 55°F on May 20?" The CLOB ask is 22¢. Your forecast says the probability is 65%. You have a €200 bankroll.
Full Kelly says bet 55% of your bankroll — €110. That feels extreme, because it is. Full Kelly is theoretically optimal but psychologically brutal and very sensitive to errors in your probability estimate.
Why use quarter-Kelly in practice
If your probability estimate is even slightly off, full Kelly can lead to catastrophic drawdowns. The practical solution is fractional Kelly — typically quarter-Kelly (25% of the full Kelly fraction).
Quarter-Kelly captures most of the growth benefit while dramatically reducing variance and the impact of model errors. It's the industry standard for professional sports bettors and quantitative traders.
PolyEdge uses quarter-Kelly with an additional 5% portfolio cap — you never bet more than 5% of your bankroll on a single market, regardless of what Kelly says.
Three rules to avoid blow-ups
- Never bet on a market with negative edge. If your probability estimate is below the CLOB ask price, Kelly outputs a negative number — meaning "don't bet."
- Use the CLOB ask, not the last-trade price. Last-trade prices on Polymarket can be hours stale. If the market has moved against you and you're using old data, your edge calculation is wrong.
- Treat correlated markets as one position. If you have edge on "Lakers win game 3" and "Lakers win series," these outcomes are correlated. Kelly assumes independent bets. Size accordingly.
Kelly and Polymarket specifically
Polymarket has two important differences from traditional betting markets:
- No house edge: Polymarket charges a 2% fee on winnings. Adjust b accordingly: if you buy Yes at 22¢ and it resolves Yes, your net payout is (1−0.22)/0.22 × 0.98. This slightly reduces the Kelly fraction.
- Markets can close early: Resolution events (elections, sports results, weather readings) are discrete. You can't exit mid-way like a stock. Size conservatively for illiquid markets.
Kelly sizing built in
PolyEdge calculates quarter-Kelly bet sizes automatically. Upload your probability estimates, set your bankroll, and get sized recommendations for every positive-edge market.
Open Free Scanner →For informational purposes only. Not financial advice. Prediction markets involve risk of loss.