Prediction markets: trading an outcome instead of a price

A contract is defined by a question, a deadline and a source that answers it. Buy the Yes side and you receive one dollar per contract if the answer is yes at the deadline, and zero otherwise. The No side is the opposite and the two always sum to one dollar, because exactly one of them pays. That constraint is what makes the market two-sided without needing a bookmaker: selling Yes at sixty two cents and buying No at thirty eight cents are the same position.

If a contract that pays one dollar is trading at sixty two cents, traders are collectively pricing the outcome at roughly a sixty two percent chance. Roughly, because fees and the cost of tying up capital until the deadline both push the price slightly below the true belief. The payoff arithmetic follows directly: pay sixty two cents, receive one dollar if right, so the gain is thirty eight cents on sixty two risked, about sixty one percent. Lose and the sixty two cents is gone in full.

Everything that makes a prediction market tradable is in its resolution rule. A precise source, a precise timestamp, and a stated answer for the boundary cases. Vague questions produce disputes, and a market that can be argued about after the fact is one whose price stops meaning anything before the fact. This is why the short-window price markets are so tightly specified: the source is a published price stream, the instant is fixed, and the tie case has an answer.

The up or down family asks whether a price finishes a fixed window above where it started, which works differently from a market on a news event with an uncertain date.

• Each window is a fixed stretch of time asking whether the price ends above where it started
• Five minute windows settle on a Chainlink TWAP stream with a thirty second lookback
• Fifteen minute windows use the same kind of stream with a sixty second lookback
• The settling number is an average, so the spot ticker can sit dollars away from it
• An exact tie resolves Up

A perp pays linearly: twice the move is twice the result, and there is no natural end to the position. A prediction contract pays a step function: being right by one cent and being right by a thousand dollars pay identically. There is no leverage to set, no funding to pay and no liquidation, because the most you can lose is already posted. In exchange you get a deadline you cannot extend and a payoff that ignores how right you were.

The useful property is that the number is backed by money. A survey costs a respondent nothing to answer carelessly; a price requires someone to fund their opinion and to be wrong in public and in cash. That does not make prediction markets accurate in any absolute sense, and thin markets with few participants are as noisy as anything else. It makes them a different kind of estimate, with a different set of biases from the alternatives.