Kelly Criterion for Prediction Markets: Stop Overbetting Your Edge
Finding a positive-EV trade is half the job. The other half is not destroying your bankroll by sizing it wrong.
Most prediction market traders overbet. They find a market they like, they feel good about it, and they put in way more than they should. When it loses — as even 70% probability trades lose 30% of the time — they are demoralized or wiped out. The Kelly Criterion exists to prevent exactly this.
Kelly is not about limiting your upside. It is about maximizing long-run growth. The math proves that betting more than Kelly reduces your long-run expected wealth, even on positive-EV trades. Overbetting is not boldness. It is a slow bleed.
The Formula
Kelly Fraction = (edge) / (odds)
For a binary prediction market contract:
f* = (p × b − q) / b
Where:
- p = your probability that YES resolves
- q = 1 − p (probability of NO)
- b = the net payout odds on a winning bet (if you buy YES at 30¢, b = 70/30 = 2.333)
- f* = the fraction of your bankroll to bet
Example: You think a market has 55% probability of YES. It is trading at 40 cents. The net odds are (1 − 0.40) / 0.40 = 1.5.
f* = (0.55 × 1.5 − 0.45) / 1.5 = (0.825 − 0.45) / 1.5 = 0.375 / 1.5 = 25%
Full Kelly says put 25% of your bankroll on this trade.
Full Kelly Is Too Aggressive
The math of full Kelly is correct in a world with perfect probability estimates and no other investments. The real world is neither.
Your 55% estimate might be 50%. Or 60%. You are not certain. Full Kelly on an uncertain estimate creates enormous variance. Two back-to-back losses — each individually expected — can crater a bankroll sized at full Kelly.
The professional standard is half Kelly at most, and often quarter Kelly. Half Kelly keeps 75% of the growth rate of full Kelly while dramatically reducing variance. Quarter Kelly is more conservative still — right for new positions, uncertain estimates, or markets where your information edge is thin.
The rule I use:
- High confidence, strong historical base rate, multiple confirming data points: Half Kelly
- Moderate confidence, mixed signals: Quarter Kelly
- New position, limited data, high uncertainty: Eighth Kelly
Eighth Kelly sounds tiny. On a 25% full Kelly calculation, eighth Kelly is 3.125% of bankroll per trade. That is still meaningful sizing. It is not a nibble — it is a professional position that does not threaten the portfolio if it loses.
Why This Matters More on Prediction Markets Than in Sports Betting
Sports bettors operate with relatively narrow lines. A −110 vig on NFL spreads means you are betting $110 to win $100. The markets are liquid and competitive. Edge is thin.
Prediction markets are different. You can find a market pricing 28 cents on something you believe is genuinely 45%. That is a 17-point gap — an edge percentage that is rare in any competitive market. When you find an edge this large, the temptation to overbet is extreme.
The Kelly math does not care about your emotional conviction. A 45% true probability at 28 cents gives a full Kelly bet of roughly 24% of bankroll. That is real money. But double it because you feel strongly, and you are in the land of ruin — where a string of losses, each individually within expected variance, wipes out your capital before the edge has time to compound.
The Bankroll Rule
Before you use the calculator, define your bankroll. Not "my net worth." Not "money I could put into this if I had to." Your prediction market bankroll is a specific, finite number you have allocated to this activity and are prepared to see move around.
Sizing 25% of a $500 bankroll ($125) is a very different decision from sizing 25% of $50,000 ($12,500). The math is the same. The practical implications are not.
Keep your bankroll separate and visible. The people who blow up on prediction markets are usually people who blur the line between their trading account and their operating cash. When you need the money, you make sizing decisions based on emotion, not math.
The Correlation Problem
Kelly assumes independence between bets. Your positions are not independent.
If you have a quarter-Kelly position on Fed holds in March, a quarter-Kelly position on Fed holds in May, and a quarter-Kelly position on 10-year yields staying above 4.5%, those three positions are highly correlated. A hot CPI print moves all three in the same direction simultaneously. Your effective exposure is much larger than any single Kelly calculation suggests.
The fix: mentally group correlated positions and apply Kelly to the group, not each trade individually. If your total macro-bearish exposure is three quarter-Kelly bets on correlated Fed outcomes, you are effectively at three-quarter Kelly on a single macro thesis. That might be fine — or it might be more than you intended.
The Practical Workflow
1. Calculate your probability estimate for the outcome
2. Check the current market price
3. Confirm you have positive EV (edge above your minimum threshold)
4. Enter bankroll, probability, and market price into the Kelly calculator
5. Apply your confidence modifier (half, quarter, or eighth Kelly)
6. That number is your maximum position size
The calculator does steps 4 and 5 in ten seconds. The discipline is in steps 1 through 3 — making sure your probability estimate is real before you size off it.
Calculate your position size now.
A good trade sized wrong is still a bad trade. Kelly is how you make sure the math works in your favor over time, not just on any single outcome.