The Kelly Criterion: How to Size Your Prediction Market Bets

Most prediction market traders get position sizing wrong. They either go all-in on high-conviction plays and blow up, or spread capital so thin that even a strong edge barely moves their P&L. The Kelly criterion solves this problem with a single formula: given your estimated edge and the market odds, it tells you the exact fraction of your bankroll to wager. Developed in 1956 and battle-tested across blackjack tables, hedge funds, and now prediction markets, Kelly remains the gold standard for bet sizing.

What Is the Kelly Criterion?

The Kelly criterion is a mathematical formula that determines the optimal fraction of your bankroll to wager on a bet with positive expected value. It was invented by John L. Kelly Jr. at Bell Labs in 1956, originally as a solution to a problem in information theory involving noisy telephone signals. The formula maximizes the long-term geometric growth rate of capital, meaning it compounds wealth faster than any other fixed-fraction strategy over repeated bets. Its core insight is deceptively simple: bet too big and you risk ruin; bet too small and you leave money on the table.

Kelly's formula gained fame when mathematician Ed Thorp used it to beat Las Vegas blackjack in the 1960s, turning a theoretical edge of 1-2% into consistent profits by sizing bets precisely. Thorp later applied the same framework to financial markets at Princeton Newport Partners, generating a 20% annualized return over 20 years with zero down years. Today, Renaissance Technologies, Bridgewater, and other quantitative firms use Kelly-derived position sizing. For prediction market traders on Polymarket and Kalshi, where binary contracts have clearly defined payoffs, the Kelly criterion is an especially natural fit.

The Kelly Formula Explained

The standard Kelly formula for a simple bet with binary outcome is expressed as a fraction of your total bankroll. In its general form, you need three inputs: the probability you assign to winning, the probability of losing, and the payout odds. The formula balances expected profit against risk of loss, always favoring the bet size that maximizes long-run growth. Here is the formula:

For binary prediction markets, the formula simplifies further. Since a winning contract pays $1.00 and you buy at the market price, the net odds are determined entirely by the purchase price. If the market price is 48 cents, your potential profit on a win is 52 cents per contract. The simplified Kelly formula for prediction markets becomes:

This version is particularly clean because p_market is just the contract price in dollars. If a contract trades at $0.48, then p_market = 0.48. The numerator (p_true - p_market) is your perceived edge, and the denominator (1 - p_market) normalizes by the potential profit per dollar risked. When f* is negative, the formula is telling you not to bet at all, or to take the other side of the contract. When f* equals zero, you have no edge.

Kelly Criterion for Binary Prediction Markets

A worked example makes the formula concrete. Suppose you are analyzing a contract on Polymarket and you believe an event has a 55% true probability of occurring. The market currently prices the YES contract at 48 cents, implying a 48% probability. You plug these values into the simplified Kelly formula:

On a $10,000 bankroll, the Kelly-optimal bet is $1,346. Your expected value per dollar invested is 14.6 cents: you pay $0.48 per contract with a 55% chance of receiving $1.00, giving an expected return of $0.55 on a $0.48 outlay. That is a 14.6% expected return per resolution. Over 100 similar bets, Kelly sizing compounds this edge into significant portfolio growth. By contrast, a flat 5% allocation ($500 per bet) would capture only 37% of the growth that Kelly delivers, while a reckless 30% allocation would increase the probability of a drawdown exceeding 50% to roughly 1 in 4.

Why Most Traders Bet Too Big

Overbetting is the most common and most costly mistake in prediction markets. In a landmark 2016 experiment by Victor Haghani and Richard Dewey, 61 finance professionals were given a coin with a known 60% edge and a $25 starting bankroll. Despite the mathematically clear advantage, 28% of participants went completely bust. The median payout was just $75, far below the $3,220 that optimal Kelly sizing would have produced. Even sophisticated traders routinely oversize when emotions and overconfidence enter the equation.

This is why experienced practitioners almost universally recommend half-Kelly, or betting half the fraction that the full Kelly formula suggests. There are three reasons this adjustment is critical for prediction market traders. First, estimation error: your edge estimate carries uncertainty, and overestimating your true probability by even 3-5 percentage points can turn a profitable strategy into a losing one. Academic research suggests reducing Kelly by 40-60% to account for estimation noise. Second, correlation: when you hold multiple prediction market positions, they are rarely independent. Macro events like a recession or a geopolitical crisis move many contracts simultaneously, meaning your effective exposure is larger than it appears.

Third, psychological tolerance: full Kelly produces approximately a 40% peak-to-trough drawdown probability, which most traders cannot stomach without deviating from their strategy. Half-Kelly, by contrast, sacrifices only 25% of the long-term growth rate while cutting variance by 50%. On a risk-adjusted basis, half-Kelly is almost always the superior choice. The math is clear: if full Kelly targets 13.5% of your bankroll, half-Kelly says 6.75%, and over hundreds of bets, your terminal wealth will be 75% of what full Kelly delivers but with dramatically smoother equity curves.

How to Calculate Kelly on Polymarket

Here is a step-by-step walkthrough using a real-world scenario. Suppose you are analyzing the "Fed Holds Rates at June Meeting" contract on Polymarket. The contract is currently trading at 64.5 cents, implying a 64.5% probability. You have built a Monte Carlo simulation using CME FedWatch data, recent CPI prints, and FOMC dot plot projections, and your model outputs a fair value of 68.2 cents.

1. Identify the market price (p_market): 64.5 cents = 0.645
2. Determine your fair value (p_true): 68.2 cents = 0.682 (from Monte Carlo simulation)
3. Calculate your edge: 68.2 - 64.5 = 3.7 cents per contract
4. Apply full Kelly: f* = (0.682 - 0.645) / (1 - 0.645) = 0.037 / 0.355 = 10.4% of bankroll
5. Apply half-Kelly (recommended): 10.4% / 2 = 5.2% of bankroll
6. Size the position: On a $5,000 bankroll, that is a $260 position, buying approximately 403 YES contracts at $0.645 each

If the event resolves YES, your 403 contracts pay $403, a profit of $143 on a $260 investment, yielding a 55% return. If it resolves NO, you lose the $260. Your expected profit per resolution is $260 multiplied by your 5.7% edge (the difference between your 68.2% fair value and the 64.5% market price, divided by the cost), which comes to approximately $14.82 in expected value. Repeating this across 50 contracts per month with similar edges produces meaningful compounding over time.

Common Kelly Mistakes

Even traders who know the Kelly formula make predictable errors in its application. Avoiding these five mistakes is the difference between a formula that grows your bankroll and one that accelerates your losses. Each mistake stems from treating Kelly as an exact prescription rather than a framework that requires calibration to real-world conditions.

1. Using full Kelly without adjustment. As discussed above, full Kelly assumes you know your true edge precisely. In practice, estimation error means full Kelly overbets by 40-60%. Always use half-Kelly or fractional Kelly (typically 0.3x to 0.5x) as your default.
2. Ignoring transaction costs. Polymarket charges no explicit trading fees, but gas costs on Polygon, spread slippage, and the opportunity cost of locked capital reduce your effective edge. A 3.7-cent edge that costs 0.5 cents in friction is really a 3.2-cent edge, which changes the Kelly fraction from 10.4% to 9.0%.
3. Not updating probability estimates. Kelly is only as good as your input probability. If new information shifts p_true from 68% to 65%, your edge shrinks from 3.7 cents to 0.5 cents and the Kelly fraction drops from 10.4% to 1.4%. Traders who set positions and forget them are implicitly betting their original estimate remains correct for weeks or months.
4. Applying Kelly to correlated bets independently. If you hold YES positions on five contracts that all benefit from a strong economy, a single macro shock can hit all five simultaneously. Treating each bet as independent inflates your total Kelly allocation. Correct practice: calculate portfolio-level Kelly or reduce individual Kelly fractions by 30-50% when holding correlated positions.
5. Treating Kelly as exact science. The Kelly criterion assumes a known edge, binary outcomes, and infinite repeated trials. Real prediction markets have multi-outcome contracts, time-varying odds, and limited resolution events. Use Kelly as directional guidance for sizing, not as a precise dollar figure to follow blindly.

The Kelly criterion is the best framework we have for translating edge into position size. But it is a framework, not an oracle. Combined with rigorous fair value estimation from tools like Monte Carlo simulation, it gives prediction market traders a disciplined, mathematically grounded approach to bankroll management that separates long-term winners from the 90% who overbet and eventually blow up.