How to find edge on a prediction market

Almost every losing prediction-market trader is losing for the same reason: they found an opinion and called it an edge. An opinion is *I think Arsenal win this*. An edge is *I think Arsenal win this 61% of the time, the market is charging me 55c, and it costs me about 1 point to get in and out — so I am being paid 5 points to take a bet I would take at even money*. The second one is arithmetic. The first one is a feeling.

Edge has exactly three parts, and you need all three. Miss any one and the number you end up with is not edge, it is decoration.

1. The market's number

On Polymarket and Kalshi this is easy, because the price *is* the probability. A contract trading at 55c pays $1 if it resolves YES, so the market is saying roughly 55%. On a sportsbook it takes one more step — American odds of −140 imply 140 / (140 + 100) = 58.3% — and that number is inflated, which is the subject of the next section.

2. Your number

This is the hard part, and there is no shortcut. What separates people who beat these markets from people who donate to them is not access to secret information — it is a repeatable process for turning public information into a number, and a record of how those numbers turned out.

A workable process looks like this:

1. 1Start from a base rate. Before you look at anything specific, ask how often this kind of thing happens in general. Home favourites of this size win about X% of the time. Incumbent parties hold about Y%. Starting from the base rate and adjusting is how forecasters stay honest; starting from the story is how they talk themselves into 80%.
2. 2Adjust on evidence you can point to. A named injury, a dated poll, a published earnings figure. If you cannot say which specific fact moved your number and by how much, it did not move your number — your mood did.
3. 3Write the number down before you look at the price. Anchoring is not a thing that happens to other people. If you price the market first, your "independent" estimate will land suspiciously close to it.
4. 4Say how wrong you might be. "61%, and I would not be surprised by anything between 54% and 68%" is a much more useful output than "61%". A wide band is not weakness; it is information about how hard to bet.

3. What it costs to be there

Every market charges you something for the privilege of having an opinion. On a sportsbook it is the vig baked into the price. On Polymarket and Kalshi it is the spread between the bid and the ask, plus fees, plus whatever slippage you eat if your size is bigger than the top of the book. That cost comes off your edge before anything else, and it is why a 2-point disagreement with the market is not a bet.

The part nobody wants to hear

Real edges are rare. If your process is producing an actionable edge on most of the markets you look at, the process is broken — you are almost certainly reading your own confidence back as signal. On a large sample of markets, the price is right far more often than it is wrong, and the correct output most of the time is a small position or none at all.

If you cannot say what would make you wrong, you have not made a forecast. You have made a prediction, which is a different and much less useful thing.

How to know if any of this is working

Track your calls against what actually happened, and score them properly. A Brier score is the standard tool: it rewards being right and punishes being confidently wrong, which is exactly the behaviour you want. Do it over at least 30 resolved calls before you conclude anything — under that, you are reading noise.