Core Concepts

At any given moment, markets are open on questions that have no ticker symbol: whether a central bank cuts rates at its next meeting, which candidate wins an election, whether a storm reaches hurricane strength before landfall. These are prediction markets. The contracts they trade settle at a fixed value once the question is answered, and until then they trade freely, so every listed question carries a running price maintained by people with money at stake. This guide explains the system behind those prices: what a prediction market is, how a question becomes a tradable contract, what kinds of events are listed, who takes the other side of a trade, and how the model compares with sports betting on one flank and conventional investing on the other.

What is a prediction market?

A prediction market is a marketplace where participants trade contracts tied to the outcome of a future event. Each market poses a question with a verifiable answer, such as whether a candidate wins a specific election, and offers positions on both sides of it: Yes and No. When the event resolves, contracts on the side that occurred settle at a fixed value, typically $1, and contracts on the other side settle at zero. Between listing and resolution, the contracts trade continuously at whatever prices buyers and sellers agree on.

The price is where the mechanism earns its name. Suppose a Yes contract trades at 34¢. A buyer commits 34¢ per contract for a claim that pays $1 if the event happens, and the participant on the other side holds a claim that pays $1 if it does not. Neither would accept those terms unless the price roughly matched their view of the outcome's likelihood, so the market price behaves as a collective probability estimate. A 34¢ price reads as roughly a 34% chance. Why that conversion works, and where it can mislead, is the subject of Prices and Probabilities. As new information arrives, participants adjust their positions and the price updates within minutes rather than on a polling or publication cycle.

The same idea travels under several names. Alongside prediction market, the research literature has called them information markets, idea futures, and event derivatives, and regulated venues list their products as event contracts. The concept is also far older than the current platforms. Speculation on papal succession appears in the historical record as early as the sixteenth century, and the Iowa Electronic Markets, a university-run research market opened in 1988, has operated continuously ever since. History of Prediction Markets traces the full arc from those origins to the modern exchanges.

Where these markets operate varies more than how they work. Some run as federally regulated derivatives exchanges, some are offered through brokerage accounts alongside stocks and options, and some run on blockchain rails where participants custody their own funds. The trading experience is broadly similar across them; the differences concentrate in access, custody, and oversight, and Venue Types maps that landscape.

How do prediction markets work?

Every market begins as a question, and much of the craft in running one lies in writing that question precisely. The listing defines what counts as the outcome, which source provides the answer, and when the answer is taken. Definitions carry enough weight that a market on whether two leaders meet before a deadline must specify what qualifies as a meeting, and a market on a company shipping a product must specify what counts as shipping. Resolution rules have to be drawn this carefully because settlement is binary. There is no partial credit between $1 and $0, so the boundary cases must be decided on paper before trading starts. How individual contracts are specified, including the way multi-outcome events are broken into binary pieces, belongs to Event Contracts.

Once the market opens, participants can take either side of the question. Buying Yes expresses the view that the event will happen; buying No expresses the view that it will not, and both sides are available at some price for as long as the market trades. Prices then move for the same reasons any market moves: information and positioning. A strong economic report, an injury announcement, a court filing, or a shift in the public conversation around an event can reprice a market quickly, and the price at any moment reflects the aggregate view of everyone willing to commit money at that level.

Holding to resolution is optional. A position can be closed at any time by selling at the current market price, which is how traders take profits early or cut losses when the picture changes; venues confirm this directly in their own documentation, as in Polymarket's help article on selling before resolution. Suppose you bought Yes at 34¢ and new information pushes the price to 60¢. Selling captures the 26¢ move per contract without waiting for the event and without carrying the risk that the market swings back. On active markets, a meaningful share of trading can be this kind of repositioning rather than positions held to the end.

When the outcome is known, the market resolves. The resolution process consults the source named in the rules, determines which side occurred, and settles every contract, paying $1 to the side that resolved true and nothing to the other. On regulated exchanges, resolution follows the contract terms filed with the venue; on blockchain venues, it typically runs through an optimistic oracle process in which a proposed outcome stands unless it is challenged within a set window. There can also be a gap between the event occurring and the market formally resolving while the named source publishes its figures or the verification process runs, and contracts often keep trading through that window at prices near the extremes. Whatever the venue, the defining property holds: the final value of the contract is decided by the event, not by any counterparty's discretion.

What can you trade on a prediction market?

Listed markets span most of public life. The categories below recur across venues, with the specific questions rotating through seasons and news cycles.

What unifies these categories is a resolution requirement. Every market needs an outcome that a named source can verify on a known date. Election results, government statistics, final scores, closing prices, and award announcements all make clean resolution sources, which is why they anchor the major categories. Open-ended questions rarely appear as listed markets, while dated, thresholded versions of the same questions do. "Will this technology succeed?" is not resolvable; "will the product reach a stated milestone by a stated date, per a named source" is.

Activity is not spread evenly across the categories. One independent analysis of several thousand markets found trading volume concentrated overwhelmingly in sports and crypto during the period it studied, with political and economic questions accounting for a far smaller share. The mix also shifts with the calendar, and an election season or a major tournament can reorder it for months at a time.

Not every position in these markets is a forecast for its own sake. Because contracts settle at a fixed value when a defined event occurs, they can offset a real-world exposure. A business hurt by a specific weather outcome or policy decision can hold Yes on that outcome so the proceeds at settlement cushion part of the loss. The CFTC's educational material presents this hedging use alongside the more common speculative one.

Who are you trading against on a prediction market?

You trade against other participants, not against the platform. The venue operates the marketplace, listing the questions, matching orders, holding the funds that back open contracts, and running settlement. It does not take the other side of your trade. That is the deepest structural difference between an exchange and a sportsbook, and the reason the money at settlement comes entirely from the participants who took the other side.

The accounting makes this concrete. Every contract has a Yes side and a No side, and the two sides together always fund the full settlement value:

$$P_{\text{Yes}} + P_{\text{No}} = \$1$$

The prices of the two sides of the same contract sum to one dollar, because whoever holds the correct side at resolution collects that combined dollar. If Yes trades at 34¢, the same claim priced from the other direction is worth 66¢. When a new contract comes into existence, a Yes buyer at 34¢ is matched with a No buyer at 66¢, the combined dollar sits in escrow, and at resolution it is paid to the side that occurred. Nothing about the outcome changes what the venue collects. In a live market the two quoted sides can drift a cent or so apart across the bid-ask spread; Order Books explains that machinery, along with how venues earn from fees on trading activity rather than from participant losses.

The question of whether the platform is the counterparty comes up often enough that venues answer it in their own documentation: Polymarket states directly that it is not the house and that trades occur between users, and Kalshi describes earning revenue from trading fees rather than from trading against its customers. An exchange on this model profits from activity rather than from outcomes, which removes the structural conflict of interest that defines a bookmaker relationship.

In practice, the party on the other side of your trade is one of three types: a participant who holds the opposite view, a market maker quoting both sides continuously and earning the spread between them, or someone exiting a position they opened earlier. All three are ordinary accounts operating under the same rules. That has a consequence worth understanding early: prediction markets are competitive. When you buy at 34¢, someone chose to sell at that price, and on active markets the seller may be better informed than the average participant. Prices tend to be sharp precisely because trading against them carelessly is expensive.

How are prediction markets different from sports betting and investing?

Prediction markets sit between two familiar models and are routinely confused with both. The structural differences are easiest to see side by side.

Compared with sports betting

A sportsbook is a counterparty business. The operator quotes a price, takes the other side of each position itself, and builds a margin into the quote, commonly called the vig, so that a balanced book earns money regardless of the result. Prices are the operator's to set and adjust, and early exit, where offered, takes the form of a cash-out figure the operator calculates. The structure is closer to a retail counter than to an exchange.

A prediction market inverts that arrangement. The venue quotes nothing and holds no position; prices come from the participants, both sides of every question are available, and exiting early is an ordinary trade rather than a discretionary offer. Sportsbook pricing conventions can also obscure the probability estimate inside a quote, whereas an event contract's price states the estimate plainly, in cents on the dollar. The full comparison, including how to convert odds formats into implied probabilities, is in Markets vs Sportsbooks.

Compared with investing

A share of stock is an open-ended claim on a business. It has no expiration, no fixed ceiling, and a value that ultimately tracks the enterprise behind it. An event contract is a closed question. It references one outcome, expires on a known schedule, and can only ever settle at $1 or nothing. That bounded structure has clean properties, since the maximum loss is the price paid and the maximum settlement value is known in advance. It also has an unforgiving one: there is no dividend, no compounding, and no holding through a drawdown in hope of recovery, because a contract that resolves against you is worth zero. Fidelity's educational treatment of prediction markets makes the same point in risk terms, listing total loss of the amount committed among the principal risks.

The nearest traditional relative is the options market rather than the stock market. Both trade instruments with expiration dates at prices that embed probability estimates, and both can lose the entire amount committed when the thesis is wrong. Event contracts are trading instruments with defined risk, not savings vehicles, and the position-sizing discipline experienced options traders apply transfers to them directly.

How accurate are prediction markets?

Accuracy is the property everything else rests on, and it has a research record going back decades. The Iowa Electronic Markets, operating since 1988, compared its election market prices against 964 contemporaneous polls across five presidential cycles and found the market closer to the eventual result about three-quarters of the time. The canonical economics survey of prediction markets reached a similar conclusion across settings, documenting market forecasts that tend to outperform moderately sophisticated benchmarks, polls and expert panels included.

The mechanism behind that record is aggregation with incentives. A market price summarizes the views of everyone trading, weighted by their willingness to commit money, and participants who are systematically wrong tend to lose capital and influence over time. The idea that dispersed groups can out-forecast experts under the right conditions was popularized as the wisdom of crowds, and a prediction market is the financial version of the aggregation step; Wisdom of Crowds examines those conditions and the ways they fail.

The record comes with real qualifications. Accuracy depends on participation, and a thin market's price can reflect a handful of positions rather than a considered consensus. Skeptics point to episodes where political markets performed poorly and to their vulnerability to manipulation when trading is thin. The accuracy itself may also have a narrower source than the aggregation story suggests, since a working paper analyzing the complete transaction history of a major venue attributes much of it to a small minority of persistently skilled traders whose profits come from the rest of the market. On that reading, a price is less a democratic average than a running record of where informed money has pushed it.

The practical way to evaluate accuracy is calibration. Across many markets, contracts priced at 70¢ should resolve Yes roughly 70% of the time, and contracts priced at 30¢ roughly 30% of the time. Calibration can only be judged in aggregate, never from a single market. Forecasters train the same property in themselves, scoring their probability judgments against outcomes over time, and Superforecasting covers that craft.

What are prediction markets used for?

The most direct use is speculation in its plain sense. You take a position where you believe the market has the probability wrong, with a defined maximum loss and a visible consensus to trade against. Everything a trader needs to disagree with is stated in the price.

A larger group uses the prices without ever trading them. Because prices read as probabilities, an open market functions as a continuously updated public forecast, and market-implied probabilities appear in news coverage and analyst commentary alongside polls and expert projections. Institutional research has examined those prices as a dataset in their own right, an input alongside surveys and economic indicators, with the caveat that "the wisdom of the crowd only works when you have a crowd", which is to say forecast value depends on the liquidity behind the price.

The contracts also serve as hedging instruments for parties with real exposure to an outcome, in the way the CFTC's explainer illustrates with weather contracts. And companies including Google, HP, Ford, and Microsoft have operated internal prediction markets on questions like launch dates and sales forecasts, on the logic that employees hold dispersed information that formal reporting channels aggregate poorly.

What none of these uses gets from the market is an explanation. A price move records that the consensus changed; it does not say why, and the contract carries no narrative alongside its number. One assessment of the field argues that the binding constraint on prediction markets' usefulness is not the supply of forecasts but the demand for them, and a probability without context is hard to act on. The context lives upstream of the price, in the reporting, data releases, and public conversation that traders react to. Measuring that conversational layer at scale, from the overall tone to the specific emotions driving it, is the territory of Sentiment Analysis.

Related guides

The guides below continue from here, one layer down.