Updated July 2026

Expected value in prediction markets

Expected value is the single idea that separates trading from guessing. It tells you what a position is worth on average, before you know how it turns out. Once you think in EV, you stop asking "will this happen?" and start asking the only question a market actually pays you to answer: "is this price wrong?"

What expected value actually measures

Expected value, or EV, is the average outcome of a bet if you could repeat it many times. The general formula weighs each outcome by its probability:

EV = (probability you win x what you gain) minus (probability you lose x what you pay)

Prediction markets make this unusually clean, because a binary contract has exactly two endings. A YES share on Polymarket pays 100 cents if the event happens and 0 if it does not, and the price you pay is quoted in cents. Plug that into the formula and almost everything cancels:

• If YES resolves true, you gain 100c minus the price per share.
• If YES resolves false, you lose the price per share.

Write your own estimated probability as p and the price as c, and the math collapses to one line: EV per share = (p x 100c) minus c. In plain terms, your expected value in cents is simply your probability estimate minus the price. That single subtraction is the whole game. If the evidence says an event is more likely than the price implies, buying YES has positive EV. If the evidence says it is less likely, YES has negative EV and the NO side is where the value sits. When your estimate and the price match, the EV is zero and there is nothing to trade.

Worked examples, in cents

Say a market prices YES at 58c and your research, base rates, polls, whatever the evidence supports, points to a true probability around 70 percent. Run the two branches:

• 70 percent of the time you win 42c: that contributes +29.4c.
• 30 percent of the time you lose 58c: that contributes -17.4c.

Net expected value: +12c per share, which is exactly 70 minus 58. Buy 100 shares for $58 and your expected profit is $12, a roughly 20 percent expected return on the capital at risk. That does not mean you make $12. It means that across many trades with this same quality of edge, the average result trends toward that number.

Now flip the inputs. Same 58c price, but the evidence honestly supports 58 percent. EV is zero. Every trade you make here costs you the spread and any fees for no expected gain, which is a slow leak, not a strategy. One more: a heavy favorite trades at 90c and your work says 85 percent. EV is 85 minus 90, or -5c per share. You will win this trade most of the time and still lose money over a career of making it. Meanwhile a longshot at 20c that the evidence puts at 30 percent carries +10c of EV per share even though it fails seven times out of ten.

The edge is the gap, not the winner

Those last two examples carry the most important lesson in this guide. Positive EV betting has nothing to do with picking winners. It comes from the gap between the evidence and the price. The 90c favorite usually wins and is still a bad buy. The 20c longshot usually loses and is still a good one. A prediction market never pays you for being right about the world. It pays you for being more right than the price, and it charges you every time you are less right than the price, even on trades that happen to cash.

This is why "I called it" is not evidence of skill. Anyone holding YES on a 90 percent favorite calls it nine times out of ten. The question that decides whether you make money over hundreds of trades is whether the prices you paid were, on average, below the true probabilities you bought. Edge lives entirely in that difference, which is also why liquid markets are hard to beat: thousands of traders have already pushed most prices close to the evidence, and the leftover gaps are small, rare, and usually found where attention is thin.

Sizing: Kelly intuition, then bet smaller

Finding positive EV is half the job. Surviving long enough to collect it is the other half, and that is a sizing problem. The classic reference point is the Kelly criterion, which for a binary contract suggests staking a fraction of bankroll equal to roughly the edge divided by the potential profit. In the 58c example with a 70 percent estimate, full Kelly works out to about 29 percent of your bankroll, and that number should scare you.

It should scare you because Kelly assumes your probability estimate is exactly correct, and it never is. Your 70 percent is really a range, maybe 62 to 75, built from imperfect information. If your true edge is half what you think and you size at full Kelly, you are systematically overbetting, and overbetting a real edge can still grind a bankroll to dust through volatility alone. The practical takeaway used by most serious traders is to keep the Kelly logic and shrink the number: a quarter to a half of the Kelly fraction, which in this example means low single digits of bankroll on a typical trade. Small sizing converts a fragile edge into a durable one. Ego sizing converts a real edge into a blown account.

Variance: positive EV still loses, often

Expected value describes the long run. The short run is noise, and the noise is bigger than most people expect. Take the +12c trade above: you still lose it 30 percent of the time. String together a run of similar positions and losing three in a row happens about once every 37 sequences, which means if you trade regularly it will happen to you, repeatedly, while you are doing everything right. Over 50 such trades you should expect a losing streak of three or more somewhere in the sample, and drawdowns of several consecutive losses are the normal texture of a winning record, not a sign the method is broken.

This matters because variance is where discipline dies. Traders abandon sound processes during ordinary cold streaks and double down during lucky heaters, exactly backwards. The fix is boring: judge yourself on the quality of the price you paid versus the evidence at the time, not on how the last five trades resolved. Resulting, judging decisions by outcomes, is the fastest way to unlearn good habits.

Why most traders lose

Put the pieces together and the failure modes become obvious. Most participants lose not because they are bad at predicting, but because they pay spread and fees to trade markets that are already fairly priced, which is negative EV by construction. Others are directionally decent but overconfident: their "70 percent" opinions are really 60 percent information dressed up with conviction, so the gap they think they are buying does not exist. Add oversized positions that turn survivable variance into ruin, and adverse selection, where the counterparty happily filling your order at this price may simply know more than you do, and the average outcome writes itself.

The traders who last do a few unglamorous things: they pass on fair prices constantly, they demand a gap large enough to survive fees and their own estimation error, they size small, and they keep score in EV terms rather than win-loss terms.

How PredictionSignal verdicts map to EV

Everything on this site is built around the gap. When a signal reads underpriced, it means the evidence our analyst panel gathered points to a true probability above the current YES price: positive EV territory for YES at the price shown, at the time of analysis. Overpriced means the evidence sits below the price, which makes NO the side with the positive gap. Fairly priced means the honest answer is that no exploitable gap showed up, and the EV framework says the correct trade is no trade. The fair-value range on each signal is our way of admitting that p is an estimate, not a fact, and the confidence label tells you how much weight the evidence can actually bear. Prices move, so a verdict is only as fresh as its date, and the methodology page explains how each one is built. All of it is education and research, not financial advice: the EV math is the lens, but the estimates, the sizing, and the decisions stay yours.

FAQ