- EV = True Probability − Purchase Price. The single most important formula in prediction market trading.
- A 37% win rate can generate 3× more profit than a 90% win rate — it depends entirely on the prices paid.
- Taker fees turn marginal positive-EV trades negative. Maker execution (post_only: true) is non-negotiable.
- Ex-ante EV is the only valid basis for decisions. Evaluating trades by outcomes is a cognitive error.
- Kelly Criterion: f* = (b×p−q)/b. Use ¼ Kelly in practice. Hard cap at 5% of bankroll per market.
- Track your Brier Score to distinguish genuine edge from variance over time.
Why Win Rate Is the Wrong Metric
Most traders think in terms of winning and losing. They celebrate high win rates and feel embarrassed by losses. This mental model is wrong — and actively expensive.
Consider two traders over 100 trades each:
- Trader A wins 90% of the time, buying YES at 92¢ on contracts with 93% true probability. EV per trade: +$0.048.
- Trader B wins only 37% of the time, buying YES at 15¢ on contracts with 30% true probability. EV per trade: +$0.15.
After 100 trades at $100 each: Trader A earns ~$480. Trader B earns ~$1,500. The 37% win rate generates three times more profit. Win rate tells you nothing. Expected value tells you everything.
The EV Formula
Expected value is the probability-weighted average of all possible outcomes. For binary prediction markets (two outcomes: $1.00 or $0.00), it simplifies cleanly:
Where p = your estimated true probability, q = 1−p, and Price = what you paid. This reduces to the practical shortcut:
If you believe true probability is 40% and the contract costs 30¢, EV = +$0.10. Over a large sample, you expect to earn 10 cents profit per contract traded.
Kalshi weather contract: will NYC exceed 85°F tomorrow? Market: 30¢ (30% implied). Your model: 40% true probability.
EV = (0.40 × 0.70) − (0.60 × 0.30) = 0.28 − 0.18 = +$0.10/contract
The Bid-Ask Spread and Fee Tax
You can never buy at the midpoint — you pay the ask. If midpoint is 30¢ but ask is 32¢, your EV is +$0.08, not +$0.10. Always calculate EV using the actual execution price.
With Kalshi taker fees, the true net EV is:
On a 15¢ contract, taker fees consume ~6.6% of capital — up to 56% of gross profit on a winning trade. A +$0.10 EV trade becomes −$0.03 after taker fees. This is why post_only: true (Maker execution) is non-negotiable.
Ex-Ante vs. Ex-Post: The Most Important Distinction
Calculated before the event from your model. The only number that matters for the trade decision.
Always $1.00 or $0.00 after resolution. Using outcomes to evaluate process is results-oriented thinking — a cognitive error that destroys systematic discipline.
A positive-EV trade that loses is not a mistake. If you estimated 40% and bought at 30¢, your process was correct. The loss is expected variance from a legitimately profitable strategy. Do not change strategy based on individual outcomes.
The Kelly Criterion: Optimal Position Sizing
Once you have a positive-EV trade, how much should you bet? Kelly maximizes long-term bankroll growth:
Where b = net odds (profit/stake) = (1.00−Price)/Price, p = win probability, q = 1−p.
In practice: use ¼ Kelly (25% of the formula output). Full Kelly assumes perfect probability estimates. They never are. Quarter Kelly gives you ~75% of the growth rate at a fraction of the ruin risk. Hard cap single-market exposure at 5% of bankroll.
| Contract Price | True Probability | EV | Full Kelly | ¼ Kelly |
|---|---|---|---|---|
| 88¢ NO | 92% | +$0.040 | 33% | 8.3% |
| 72¢ NO | 80% | +$0.080 | 22% | 5.5% |
| 30¢ YES | 40% | +$0.100 | 19% | 4.8% |
| 5¢ YES | 4% | −$0.010 | 0% (do not bet) | — |
The Brier Score: Measuring Real Edge
A string of profits could be skill or luck. The Brier Score separates them:
Lower is better. Perfect calibration = 0. Random guessing = 0.25. Target below 0.15. Track it every session. A declining Brier Score over time means your edge is real and improving. A flat score despite profits means variance is carrying you — reduce position sizes.
Frequently Asked Questions
On Kalshi as a Maker, fees are roughly 0.44¢/contract at 50¢. Your minimum EV threshold should comfortably exceed this — practical minimum is +2¢ to +3¢. Below this, model error can flip the trade negative. Higher-variance event types warrant higher minimums.
Kelly adapts to edge size and is theoretically optimal. Fixed fractional (e.g. always 2% of bankroll) is simpler. If you have fewer than 50 trades of Brier Score history, use fixed fractional at 1–2%. Move to ¼ Kelly once you have demonstrated calibration.
Not based on 10 trades. Even a 70% win rate has a 0.003% chance of 10 consecutive losses. Before changing anything: is your Brier Score deteriorating? Are your probability estimates drifting from outcomes systematically? If no, it is likely variance. Changing strategy based on variance is results-oriented thinking and will destroy your edge.