The Kelly Criterion is a position-sizing formula that calculates the mathematically optimal fraction of your bankroll to wager on any trade, given your estimated edge and the market odds — maximizing long-run capital growth while avoiding the ruin risk that comes from overbetting.
TL;DR: Key Takeaways
- The Kelly Criterion formula is: f* = (bp – q) / b, where b is the net odds, p is your estimated probability of winning, and q is the probability of losing (1 – p).
- Most experienced prediction market traders use fractional Kelly (25–50% of full Kelly) to reduce variance without sacrificing much long-run growth.
- Kelly only works as well as your probability estimates — edge estimation is the hardest and most important part of the system.
- Overbetting is more dangerous than underbetting: betting twice the Kelly optimal can actually result in negative expected growth despite a positive edge.
- On platforms like Kalshi and Polymarket, Kelly sizing should be adjusted for low-liquidity markets where large positions move the price against you.
- Tracking your actual win rate vs. estimated win rate over time is the most reliable way to calibrate your Kelly inputs.
What is the Kelly Criterion and why does it matter for prediction markets?
Originally developed by Bell Labs mathematician John L. Kelly Jr. in 1956 (published in the Bell System Technical Journal), the Kelly Criterion was designed to solve a specific problem: given a repeatable bet with a known edge, how much of your bankroll should you risk each time to grow wealth as fast as possible without going broke?
Prediction markets are nearly a perfect environment for Kelly-style thinking. Every contract has a binary outcome, a market-implied probability, and a fixed payout structure. If you believe the market is mispriced — that an event is more or less likely than the price suggests — Kelly tells you exactly how aggressively to act on that belief.
The core formula is simple:
f* = (bp – q) / b
- f* = the fraction of your bankroll to bet
- b = the net odds received on the bet (e.g., if you buy YES at $0.60 on a binary market, b = 0.667, since you win $0.40 on a $0.60 stake)
- p = your estimated probability the event occurs
- q = 1 – p, the probability it does not occur
How do you calculate Kelly Criterion on Kalshi or Polymarket?
Let's use a real example. Suppose a Kalshi contract is priced at 55¢ for YES on "Will the Fed cut rates in June 2026?" — implying a 55% probability. You've done your research and believe the true probability is 68%.
- b = (1 – 0.55) / 0.55 = 0.818 (you win $0.45 for every $0.55 risked)
- p = 0.68
- q = 0.32
Plugging in: f* = (0.818 × 0.68 – 0.32) / 0.818 = (0.556 – 0.32) / 0.818 = 0.289, or about 28.9% of bankroll.
That's a large bet — and this is where most newcomers get into trouble. Full Kelly is theoretically optimal but practically brutal. A string of losses, even with a genuine edge, can wipe out a substantial portion of your account before mean reversion kicks in.
Why do most traders use fractional Kelly instead of full Kelly?
The volatility of full Kelly is the reason nearly every professional — from quantitative hedge funds to poker players — uses a fractional approach. Half-Kelly (50% of the full Kelly fraction) reduces variance by roughly 75% while sacrificing only about 25% of the theoretical growth rate. Quarter-Kelly is even more conservative and still maintains most of the compounding advantage over flat betting.
In the Fed rate example above, half-Kelly would mean sizing at 14.5% of bankroll instead of 28.9%. Quarter-Kelly drops that to 7.2% — still a meaningful position, but survivable if your probability estimate turns out to be off.
The practical rule of thumb used by experienced prediction market traders: use full Kelly only when you have strong historical calibration data supporting your edge estimate. Default to half-Kelly or quarter-Kelly when you're working in less-familiar market categories.
What happens when your probability estimates are wrong?
This is the hidden risk in Kelly-based systems. The formula is only as good as your p estimate. If you consistently overestimate your edge — a very common bias — full Kelly will systematically overbet and drag down your long-run returns worse than if you'd never used Kelly at all.
Research on forecasting accuracy (Good Judgment Project calibration data) consistently shows that even skilled forecasters tend to be overconfident in high-confidence ranges (80–95% predictions). This means you should apply a systematic discount to your edge estimates, especially on events where you feel highly certain.
A practical calibration habit: track every prediction market position you enter, record your estimated probability at entry, and compare it to your actual win rate over 50+ trades in each category. If your "70% confidence" trades are winning only 58% of the time, your Kelly inputs need a correction factor of roughly 0.58/0.70 = 0.83 applied to all p estimates in that category.
For a deeper framework on managing this estimation risk, see our Prediction Market Risk Management: Complete 2026 Guide.
How does liquidity affect Kelly sizing on prediction markets?
On thin markets — Kalshi contracts with under $10,000 in open interest, or low-volume Polymarket events — full Kelly sizing creates a second problem beyond variance: market impact. A large order in a thin market moves the price against you, which means your effective entry price is worse than the price you calculated your edge against, and your real Kelly fraction should be lower.
The practical fix: on any market where your Kelly-optimal position would represent more than 2–3% of total open interest, reduce your size by 30–50% and enter in tranches rather than a single order. This preserves most of your edge without degrading the market price against yourself.
How does Kelly Criterion work across a portfolio of prediction market positions?
When you hold multiple simultaneous positions — which is standard practice — individual Kelly fractions need to be scaled down to avoid total portfolio over-leverage. A simple rule: if your individual Kelly fractions across all open positions sum to more than 25–30% of bankroll, scale each position proportionally until the sum falls within that range.
More sophisticated traders use a correlated Kelly framework, where positions on related events (e.g., multiple Fed-related contracts, or several markets all tied to the same election outcome) are treated as a single combined bet rather than independent wagers. This prevents the false diversification of holding five positions that all lose together.
For a complete framework on building a multi-position portfolio, see our guide on Mean Reversion in Prediction Markets, which covers how statistical signals interact with position sizing across correlated events.
What are the most common Kelly Criterion mistakes in prediction markets?
- Using market price as your probability estimate. The market price is the consensus, not your edge. Kelly requires your independent probability assessment.
- Applying Kelly to single-trade analysis. Kelly is a long-run system. Judging it on any individual trade outcome misses the point entirely.
- Ignoring the minimum edge threshold. If your estimated edge is less than 3–5%, transaction costs and slippage on most platforms will eliminate it. Only apply Kelly when you have a meaningful, defensible edge.
- Betting full Kelly on correlated positions. As described above, this multiplies your risk without proportionally multiplying your edge.
If you're new to the mechanics of how prediction markets work before applying any sizing system, our How to Trade Prediction Markets: Complete Beginner's Guide covers the foundational concepts.
Frequently Asked Questions
What is the Kelly Criterion in simple terms?
The Kelly Criterion is a formula that tells you what percentage of your bankroll to bet on a trade to maximize long-run growth. It balances aggressiveness — betting enough to compound gains — against ruin risk from overbetting when you're wrong.
Is the Kelly Criterion legal to use on Kalshi or Polymarket?
Yes — Kelly Criterion is simply a position-sizing strategy, not a platform feature. You calculate the optimal fraction yourself and enter orders manually or through any supported API. There are no restrictions on using mathematical sizing frameworks on any regulated prediction market platform.
How much of my bankroll should I risk per prediction market trade?
Most experienced prediction market traders risk between 2–15% of bankroll per position, using half or quarter-Kelly as their upper bound. Anything above 20% per trade introduces severe drawdown risk that can permanently impair your account even with a genuine edge.
Does Kelly Criterion work for low-probability prediction market contracts?
Yes, but with extra caution. Long-shot contracts (under 15% probability) are where edge estimation error is most dangerous — a small probability miscalculation produces a large change in Kelly fraction. Use quarter-Kelly or less for contracts priced below 20¢ until you have strong calibration data in that range.
What's the difference between full Kelly and fractional Kelly?
Full Kelly bets the mathematically maximum growth-rate fraction each time, but with high variance — bad runs can cut your bankroll dramatically before recovery. Fractional Kelly (half or quarter) bets a portion of the full amount, reducing variance significantly while retaining most of the compounding advantage over flat betting.
Can I automate Kelly Criterion position sizing on prediction markets?
Yes — platforms like Kalshi offer API access that allows programmatic order entry. The main challenge is automating the probability estimation step, which requires a reliable model or data feed for each market category. Tools like Prevayo provide analytics frameworks that can support this kind of systematic, data-driven approach to sizing and tracking positions across markets.
The Kelly Criterion is the closest thing prediction markets have to a mathematically proven optimal strategy — but it demands honest, calibrated probability estimates and the discipline to stick to fractional Kelly when uncertainty is high. Start with half-Kelly, track your calibration obsessively, and let the math compound over time. Platforms like Prevayo can help you track edge estimates, monitor position sizing across your portfolio, and identify where your probability estimates are drifting from reality — turning Kelly from a formula into a systematic edge.