Risk Management for Prediction Market Traders: Binary Math, Position Sizing and Kill Criteria

Prediction market risk starts with a simple fact: a binary position can lose 100% of what you put into it. There is no coupon, no dividend and no residual value in a losing share. That makes position sizing more important than prediction. A trader who is right 60% of the time can still go broke by risking too much on each bet.

The first step is to stop thinking in "how much can I win" and start thinking in "how much can I lose." A Yes share bought at 35 cents returns 65 cents if it wins and loses 35 cents if it does not. The gain is 186% of the stake; the loss is 100%. Those are the only two outcomes, before fees and before any early exit.

The arithmetic of a binary bet

If you buy at price p, your profit per share when the event happens is 1 - p, and your loss per share when it does not is p. The break-even probability, before fees, is exactly p. At 35 cents, you need the true probability to be above 35% to have a positive expected value.

Fees raise the bar. On Polymarket, the taker fee is shares × fee rate × price × (1 - price). For 100 shares at 35 cents in a politics market with a 4% rate, the fee is 100 × 0.04 × 0.35 × 0.65 = $0.91. Your total cost is $35.91, so the break-even probability is 35.91%, not 35%.Polymarket fees

That difference sounds small. It is not, when you trade often. A strategy with a 1% edge per trade can be wiped out by a 1% fee. Always compare your estimated probability with the all-in break-even, not the quoted price.

Position sizing: Kelly and the error bar

The Kelly criterion gives the bet size that maximizes long-run growth when you know the true probability. For a binary bet with net odds b, the formula is f = (b × p - q) / b, where p is your estimated win probability, q is 1 - p, and b is (1 - price) / price.Kelly

Take a $10,000 bankroll, a price of 35 cents and an estimated probability of 42%. The net odds are b = 0.65 / 0.35 = 1.857. The full Kelly fraction is (1.857 × 0.42 - 0.58) / 1.857 = 0.108, or 10.8% of the bankroll, about $1,077.

Most traders should not bet full Kelly, because the estimate is uncertain. If the true probability is 38% rather than 42%, full Kelly is an overbet that increases the risk of a deep drawdown. A common compromise is half Kelly, about 5.4% or roughly $540, and a hard cap below that.

Suppose you set a 2% risk cap, or $200. With a 35-cent price and a 0.91-cent fee per share, your all-in cost is 35.91 cents per share. Divide $200 by 0.3591 and you get about 557 shares. If you win, those shares redeem for $557, a profit of $357. If you lose, you lose $200, or 2% of the bankroll. The asymmetry is the point: the payoff is large, but the loss is bounded and small relative to capital.

A 2% cap is a heuristic, not a law. It is useful because it keeps a single bad estimate from becoming a portfolio-level event. The CFTC's investor guidance makes the broader version of the same point: trade only with risk capital, monitor positions and understand how fees affect returns.CFTC explainer

Correlated markets are one position

The most common sizing mistake is to treat correlated markets as independent. If you buy Yes on a candidate in three states, you have not made three 2% bets. You have made one bet on the candidate, sized at 6% of the bankroll, because the same national swing can move all three markets together.

The same applies to a party winning the presidency, Senate and House, or a company beating earnings and raising guidance. The prices differ; the drivers overlap. Add the exposures before deciding the total is acceptable.

Correlation can also hedge: a No position in one outcome can offset a Yes in a related outcome. But if both depend on the same source, date or rule, they can fail together.

A practical rule: write down the one sentence that would make all your open positions lose at once. If that sentence is plausible, you are more concentrated than your position count suggests.

Drawdown math and kill criteria

Losses compound. Ten consecutive losses at 5% of the bankroll leave you with 0.95 raised to the tenth power, or about 59.9% of the starting capital, a drawdown of 40.1%. To get back to even, you need a gain of 67%, not 40%.

At 2% per loss, ten losses leave about 81.7%, an 18.3% drawdown; the recovery is 22.4%. Smaller bets keep recovery achievable.

That is why kill criteria matter. A rule such as "stop trading for two weeks after a 15% drawdown" is not an admission of failure. It is a circuit breaker that prevents a bad week from becoming a bad year. Other useful rules:

• A daily loss limit (3% of the bankroll) and a maximum position size (2% per independent idea).
• A rule against increasing size after a loss to "win it back."
• A rule against holding a position through a resolution you do not understand.
• A scheduled review after every 20 trades, comparing realized results with your estimated probabilities.

The last one is the most important. If your 40% estimates are winning only 30% of the time, the problem is not your bet sizing; it is your model. Kelly will not save a trader with a biased probability estimate.

A worked example, end to end

You have $10,000. A politics market trades at 35 cents. You estimate the true probability at 42%, based on polling and a model you trust. The market is thin, so you plan for slippage.

Your maximum risk is 2% of the bankroll, or $200. At an all-in cost of about 35.91 cents per share, you buy 557 shares. You split the order into two limits: 300 shares at 35 cents and 257 at 36 cents. With fees, the actual risk is about $203, a small overshoot you accept.

If the event happens, you receive $557 and earn about $354 after fees, a 175% return on the amount at risk. If it does not, you lose about $203. If the price rises to 45 cents, you sell for about $250.65, pay $5.51 in fees, and net $245.14, a profit of about $42, or 21%.

The point is that exit, fees and slippage belong in the plan before entry. A 10-cent win can be a modest return after costs.

The risks the math does not capture

Kelly assumes you know the probability. You do not. The polls, the base rate or the market may know something you do not, which is why fractional Kelly and hard caps exist.

The math also assumes you can exit at the quoted price. In a thin market, you cannot. A stop-loss at 25 cents may fill at 20 cents if the book gaps. Resolution risk can override every price-based rule: a position can be profitable on the screen and still lose at settlement because the rule was ambiguous.

Finally, cheap longshots deserve extra scrutiny. Research on Polymarket found that purchases below 10 cents lost 19.3 cents per dollar, while purchases at or above 90 cents earned 0.83 cents, a pattern consistent with a favorite-longshot bias.Cardozo and Rivero-Wildemauwe A 2-cent contract is not a lottery ticket with a known 2% chance; it is a price that may embed a behavioral premium. If you buy it, demand a larger edge, not a larger position.

The rule that survives

If you remember one rule, make it this: size every position by the amount you lose if the event does not happen, not by the amount you make if it does. That single change turns a portfolio of exciting guesses into a portfolio that can survive being wrong.

Prediction markets reward information, but they punish overconfidence. The traders who last are not the ones with the best single call. They are the ones who are still there after the bad call.