How prediction markets work

If you have read what a prediction market is, you know the price reads as a probability. This is where that comes from.

The contract

Every question is a contract that pays out a fixed amount — call it 100 units — if the event happens, and zero if it does not. Nothing in between. That binary payout is what makes the price interpretable.

Suppose a contract trades at 68 units. Buying one costs 68 and returns 100 if the event occurs, so you gain 32 when right and lose 68 when wrong. Those numbers only balance out over many attempts if the event happens about 68% of the time. So the price and the probability are the same number. That is the entire trick.

How the price moves

Nobody sets the price. It is wherever buyers and sellers currently agree, so it moves when the balance of opinion moves. News lands, some participants update, they act on it, and the price adjusts within seconds.

This is why a market’s history is often more interesting than its current number. A question sitting at 40% tells you what the crowd thinks now. A question that was at 12% three weeks ago and is now at 40% tells you something changed — and the shape of that move usually points at what.

Resolution: the part that matters most

A question is only as good as its resolution criteria. Every well-written market specifies in advance what source settles it and what edge cases mean, because the arguments always happen at the edges. Does a candidate “take office” if they are sworn in on an interim basis? Does a film “release” if it goes straight to streaming?

Ambiguity here is the most common way a market goes bad. When you read one, check the deadline and the settlement condition before you look at the price — plenty of surprising-looking numbers turn out to be perfectly sensible once you notice what the question actually asks.

Liquidity, and why thin markets lie

Liquidity is how much activity sits behind the price. A heavily participated question is hard to move: a single confident participant barely shifts it, so the number reflects a genuine consensus.

A thin one is the opposite. Very little activity can swing it several points, which means the price may reflect one person’s conviction rather than a crowd’s judgement. Thin markets are not useless, but treat their precision with suspicion — the difference between 20% and 26% is noise there, where on a busy question it would be meaningful.

Where the mechanism breaks down

Long-dated questions drift toward the middle. Committing to an outcome two years out ties up whatever you staked for two years, so prices on distant events tend to be less sharp than near-term ones.

Very unlikely events sit above their true odds. Prices near 1–2% are noisy, partly because the smallest increment is a large fraction of the price and partly because people systematically overpay for long shots.

Popularity is not probability. On questions with strong emotional pull — particularly political ones — prices can lean toward what participants want rather than what they expect.

Practising without the stakes

All of this is learnable, and the cheapest way to learn it is to make calls and watch how they land. That is what PolyGround is for: the same question format and the same live probability movements, played with virtual coins that hold no monetary value and cannot be purchased or cashed out. You build the instinct for reading a number without needing money on the line.

Next: how to read the odds, or browse live questions by topic.