RESEARCH / 01
How Market-Implied Probability Forms: Price Formation, Information Loss, and Decision Boundaries
Abstract
Binary event contracts typically trade at prices between 0 and 1. A contract priced at $0.37 is therefore often interpreted as saying that "the market assigns a 37% probability to the event occurring." This conversion is convenient, but it can obscure a basic fact: 37% is first a price formed through trading, and only then is it given a probability interpretation.
Prices begin with individual judgments, pass through capital constraints, order submission, trade matching, and data presentation, and ultimately become the percentages on a screen. At each stage, some information is filtered out or compressed. Views that do not enter trading do not enter the price; capital size changes the influence of different judgments; the bid–ask spread and market depth determine whether quotes are executable; and the contract's resolution and settlement rules determine which event the market is actually trading.
This article follows that path of price formation. It first explains why a binary contract's price can be interpreted as a probability, then examines the order book to see how that probability is formed. It next analyzes the information lost during price formation and, finally, the context required when market-implied probabilities enter automated decision systems. The article argues that market-implied probability is a conditional price signal. Its value depends not only on the percentage itself, but also on who formed it, at what cost, and under which rules.
1. From Price to Probability: The Basic Logic of Binary Contracts
Suppose there is a future event whose outcome can be verified. A binary contract is issued on that event, with only two possible settlement outcomes:
Let X denote the event outcome: X=1 if the event occurs, and X=0 if it does not.
Here, X represents the contract's settlement value. If the event occurs, the contract pays $1; if it does not, the contract pays $0.
Suppose a trader buys one contract for q dollars and believes that the probability of the event occurring is p. If the event occurs, the trader receives $1 and earns a net profit of 1−q. If it does not, the contract becomes worthless and the trader loses the q dollars already paid.
The expected profit from buying one contract is:
Where:
- E[R] is the expected profit from the trade;
- p is the trader's subjective probability that the event will occur;
- q is the contract's current price.
Equation (1) establishes the relationship between an individual judgment and the market price. When p exceeds q, the trader considers the contract underpriced and finds buying attractive. When p is below q, the trader is more likely to sell or take the opposite side.
Suppose a trader assigns a 60% probability to the event occurring, while the contract is priced at $0.50:
Ignoring transaction costs for now, this trade has a theoretical expected profit of $0.10. Another trader who assigns only a 40% probability to the event would reach the opposite conclusion at the same price. Their judgments enter the market as orders on different sides.
Only under conditions such as a clearly specified unit payout, clear resolution and settlement rules, the ability to enter and exit trades freely, approximately risk-neutral participants, budget and position constraints that are not dominant, and sufficiently low trading frictions might the marginal price approach some measure of central tendency in participants' beliefs. Only under stronger structural assumptions can it be interpreted approximately as the average belief. Even then, risk aversion and the distribution of beliefs may create systematic bias [1–3]. A price of $0.37 can therefore serve only as a probability proxy under these conditions.
This derivation states the conditions under which a probability interpretation is valid; it does not prove that any particular price must be accurate. In actual trading, p cannot be observed directly, and participants differ in their tolerance for risk. The resulting price can show only which judgments were successfully translated into orders and where those orders resulted in trades.
Probability accuracy also requires long-run calibration checks. An event assigned a 70% probability may or may not occur. A single outcome determines the contract's settlement, but cannot establish whether 70% was a reasonable assessment. Only when, across many comparable events, approximately 70% of the events priced at 70% actually occur does that set of probabilities exhibit good calibration.
Prices therefore take the form of probabilities, while the calibration record determines how much trust that form deserves.
2. Probability Is Formed in the Order Book
An individual's probability judgment does not appear directly on a market page. It must first become a specific order.
In an order-book market, buyers submit the prices they are willing to pay and sellers submit the prices they are willing to accept. Let bₜ denote the current highest bid and aₜ the current lowest ask. The difference between them is the bid–ask spread:
Here, sₜ is the bid–ask spread at time t.
Suppose the highest bid is $0.36 and the lowest ask is $0.38:
The two sides are currently only $0.02 apart. When a buyer accepts the $0.38 ask, or a seller accepts the $0.36 bid, a trade occurs and produces the latest trade price.
The probability displayed on a market page may come from the latest trade price or from the midpoint of the bid and ask. If the midpoint is used, the value at time t is the quote midpoint—a probability proxy:
Here, p̂ₜ is the market probability estimate at time t. The hat above p indicates that it is an estimate.
Substituting the preceding quotes:
The page therefore displays 37%.
The calculation is simple, but the interpretation cannot stop there. The latest trade price records where the most recent trade took place, yet that trade may have occurred hours ago and may have been very small. The quote midpoint reflects the center of the current bid and ask boundaries, but does not guarantee that any participant is willing to trade at that midpoint.
Suppose another market has a highest bid of $0.20 and a lowest ask of $0.54. Its quote midpoint is also $0.37:
Both markets display 37%, but they describe entirely different market conditions.
In the first market, the two sides are only two percentage points apart, indicating relatively clear agreement in the quotes around 37%. In the second market, the two sides are 34 percentage points apart. Here, 37% is merely the mathematical midpoint between two distant quotes; the market has not yet formed a stable price.
Market depth further changes the meaning of 37%. Suppose there are only $10 worth of sell orders near $0.38. A slightly larger buy order could push the price to $0.45. In that case, 37% describes only the marginal price before a very small amount is traded. It does not establish that substantial capital is willing to bear risk near 37%.
Price, spread, and depth describe the market together. Showing only the price is like giving a thermometer reading without explaining whether it has just been updated, how large its measurement error is, or the environment in which it was placed.
3. From Judgment to a Number: Where Information Is Lost
The path from an individual judgment to an on-screen percentage passes through at least four stages: judgments enter the market, capital becomes orders, orders produce trades, and real-world events are resolved and settled according to rules. Each stage retains some information while excluding other information.
3.1 Judgments That Do Not Enter the Market Do Not Appear in the Price
A market can aggregate only the information held by those who actually participate in trading.
Some people may understand an event better but remain outside the market because they lack access to trading, have insufficient capital, face excessive operational costs, or do not trust the resolution and settlement rules. Their judgments do not enter the price automatically.
The number of participants is also not a direct measure of information diversity. Suppose one hundred traders all rely on the same public news item. They submit one hundred orders, but the underlying information comes from a single source. Another market may have only ten active traders, each with distinct knowledge of technical progress, supply-chain changes, policy positions, and capital flows. The latter market has fewer participants, but may have a richer information structure.
The price therefore aggregates information that has entered the market through participants willing to bear risk. Knowledge that has not yet been traded may still exist outside the market.
3.2 Judgments That Enter the Market Do Not Receive Equal Weight
Markets bring information together through capital, rather than giving each participant an equally weighted vote.
Participants differ in wealth, risk tolerance, and position limits. Even if nine participants each believe the event has a 60% probability of occurring, if each is willing to commit only $10 while another participant who assigns it only a 30% probability is willing to commit thousands of dollars, the latter may still exert greater influence on the final price.
Capital weighting can sometimes improve price quality. Participants with more reliable information or stronger convictions are generally willing to take greater risks. But capital size can also reflect factors unrelated to forecasting ability, such as differences in wealth, hedging needs, or short-term liquidity arrangements.
An investor may buy an event contract because they consider the event more likely, or because the event would cause losses elsewhere in their portfolio. In the latter case, the contract serves an insurance function. The buy order contains both a probability judgment and a risk-management need, and the market cannot fully separate the two from the executed trade.
Whether a trader is willing to correct a price also depends on cost. Let p be the trader's subjective probability, q the market price, and c the total cost of completing the trade. For buying to offer a positive expected net benefit, the following condition must hold:
Here, c is measured for the same contract and in the same settlement unit as p and q. It may include fees, the cost of tying up capital, expected price impact, and settlement risk. If q already uses an executable bid or ask price, the bid–ask spread is already reflected in q and should not be counted again in full within c.
Equation (4) means that small errors in a market may be identified without being traded away. As long as the price discrepancy is insufficient to cover the cost of correcting it, a rational trader may choose not to act. Market precision therefore has a tolerance band determined by trading costs.
3.3 Trade Prices Compress Disagreement in the Order Book
Once a trade occurs, the page typically retains just one price. That number compresses information previously distributed across the order book.
The same $0.50 price may come from a market in which most participants place the probability near 50%, or from two groups with opposing views: one firmly believes the event will occur, while the other firmly believes it will not. These markets may have the same price but entirely different distributions of opinion, price stability, and risks of future volatility.
Price alone cannot show the degree of disagreement among participants or indicate how much a new order would move the market. The bid–ask spread, quoted size, trading volume, and position concentration are therefore more than supplementary data alongside the price; together, they determine the quality of the price signal.
3.4 Real-World Events Must Also Pass Through Resolution Rules
What market participants ultimately trade is determined by the contract's resolution and settlement rules.
A contract may ask whether a technology will be "officially released" before the end of the year. In ordinary language, a "release" might include a public demonstration, limited testing, open registration, or full commercial availability. If the rules recognize only an official version available to the public, the market is actually trading the probability that the technology will meet the specified resolution conditions before the deadline.
Traders must assess both what will happen in the real world and whether that outcome will be recognized under the resolution rules. Contract risk arises at least from uncertainty about the event and uncertainty about how the resolution rules will be interpreted. These risks may interact and cannot be separated using price alone. The following relationship is only a conceptual model:
The clearer the resolution conditions, the more readily the price corresponds to a consistently defined event. The more ambiguous those conditions, the more rule-related risk enters the price. Equation (5) does not assume that the three risks are additive, mutually independent, or directly estimable from a single price.
This allows a conceptual decomposition of the difference between the implied probability and a reference probability. Let pₜ* denote the reference probability used for analysis at time t, and p̂ₜ the probability displayed by the market:
Where:
- εₜᴹ is the deviation attributable to market microstructure;
- εₜᴾ is the deviation attributable to participant composition;
- εₜᴿ is the deviation attributable to risk preferences and capital constraints;
- εₜˢ is the deviation attributable to resolution and settlement rules;
- εₜᴰ is the deviation attributable to data calculation and presentation;
- εₜⁱⁿᵗ is the interaction residual left by the combined effects of these factors.
Equation (6) uses a definitional symbol to organize the observed difference into a framework for locating sources of deviation; it is not a method for directly calculating a so-called "true probability." In most real-world events, pₜ* cannot be observed directly. The error terms are not assumed to be independent, uniquely identifiable, or exhaustive of all sources, and a market snapshot at a single point in time cannot estimate them separately. The equation indicates only the stages at which a market probability may be distorted.
These deviations also interact. Ambiguous resolution rules may discourage professional capital from participating; insufficient participation reduces market depth; and insufficient depth amplifies the effect of individual orders on the price. The page can still display a percentage with decimal precision, but precision in the number's presentation does not imply equally precise underlying information.
4. When Probability Enters a Machine: Context Becomes an Execution Condition
When people see a market probability, they can open the order book, check trade timestamps, read the resolution and settlement rules, and draw on experience to assess the number's reliability. A machine can process only the data actually supplied by an interface. Information missing from that interface is effectively nonexistent to an automated system.
This turns the context around a market probability from a reading aid into an execution condition.
Suppose two systems both read "the event has a 60% probability of occurring." The first figure comes from a market with continuous trading, a narrow spread, sufficient depth, and clear rules. The second comes from a small trade several hours ago, while the current order book is almost empty.
If the interface returns only 0.60, the two signals look identical to a machine. A trading program may adjust positions on that basis, a risk system may calculate exposure, and an agent may treat it as an external assessment of a policy, technology, or asset-related event. Differences in market conditions are then hidden behind the same input format.
To assess whether an implied probability can be used for decisions, this article defines the following minimum set of market-state information:
Where:
- Sₜ is the market state used to assess probability quality at time t;
- p̂ₜ is the current implied probability;
- sourceₜ is the source used to calculate the probability;
- bidₜ is the current highest bid;
- askₜ is the current lowest ask;
- depthₜ is the market depth around the current price;
- volumeₜ is the amount traded within a specified time window;
- recencyₜ records when quotes and the latest trade were updated;
- concentrationₜ is the concentration of trading volume or positions;
- resolutionₜ contains the resolution source, deadline, and dispute status.
This is a minimum disclosure vector proposed by this article, not an established industry standard. p̂ₜ, bidₜ, and askₜ are normalized to [0,1] using the settlement unit per contract. Both depthₜ and volumeₜ must specify their units of measurement and observation windows. recencyₜ is measured as the time elapsed since the latest valid quote or trade. concentrationₜ must state whether aggregation is by address, entity, or source of funds.
Equation (7) does not change the original probability; it restores the market context lost when information was compressed into a price.
Automated systems can use these fields to establish usage rules: reduce a signal's weight when the bid–ask spread widens, mark a probability as stale when trading has been inactive for a long time, limit execution size when market depth is insufficient, raise manipulation-risk alerts when positions are overly concentrated, and pause automated decisions when resolution is disputed.
The same probability can therefore receive different weights in a decision. A figure of 60% is no longer treated as an unconditional conclusion, but as an estimate accompanied by its source, freshness, and market quality.
A Probability Interface Must Provide More Than Probability
Valflux therefore proposes a data principle for markets and agents: an implied probability should be published as an object that includes market state. In addition to returning p̂ₜ, an interface should provide the source of the price calculation, bid and ask quotes, market depth, the latest trade timestamp, participant concentration, and resolution and settlement status. Human researchers can then judge whether 37% comes from stable trading or a thin order book, while automated systems can turn those fields into weights, position limits, and pause conditions.
This proposal focuses on preserving key evidence from the probability-formation process. Valflux aims to retain that evidence so that different-quality estimates of 37% receive different treatment in decisions. When spreads widen, trades become stale, or resolution is disputed, agents should reduce the signal's weight. Only when depth is sufficient, sources are transparent, and historical calibration is stable is a probability suitable for higher-risk automated execution. Sₜ is therefore both a description of market state and the minimum informational responsibility that probability infrastructure owes downstream users.
The above describes Valflux's research and protocol-design proposals. Which fields are already available in existing interfaces and which remain at the design stage must be checked individually against the current product version in the formal publication. Intended functionality must not be presented as already implemented.
5. Decision Boundaries: What Can a Market Price Establish?
What a market-implied probability directly indicates is the price at which the capital currently participating in trading is willing to bear the risk of a particular event contract's settlement outcome.
It can reveal the aggregate result of information entering the market, but cannot establish that participants possess all relevant information. It can record price changes after news appears, but cannot on its own establish a causal relationship between the news and the price. It can show the current marginal trade price, but cannot reconstruct every participant's belief distribution from a single number.
An implied probability also has an explicit time dimension. p̂ₜ at time t reflects only the information that had entered the market at that point. Once a new policy document, technical development, change in capital behavior, or resolution dispute emerges, the old price no longer represents the new market state.
Market probability is therefore suitable as one part of the evidence for a decision. It can be used alongside fundamental data, historical benchmarks, expert models, and other independent signals. It can receive greater weight when market conditions are transparent, resolution criteria are clear, trading depth is sufficient, and historical calibration is good. Its use needs to be constrained when the market is thin, the price is stale, participation is highly concentrated, or rules are disputed.
Such boundaries do not diminish the value of market prices. On the contrary, a signal can enter real decisions reliably only when its users understand what it can and cannot answer.
6. Conclusion
Market-implied probability emerges along a continuous path: individuals form judgments; judgments become orders; orders produce trades subject to capital and transaction-cost constraints; trade prices are processed into percentages; and real-world outcomes are ultimately determined under resolution rules.
Prices can be interpreted as probabilities because traders act on the difference between their subjective judgment p and the market price q. Prices cannot be treated directly as objective probabilities because not every judgment can enter the market, and those that do enter are not equally weighted. The order book's spread, depth, and trade timestamps continue to affect price quality, while resolution rules determine which event the probability actually refers to.
A standalone percentage preserves only the final output of this entire formation process. For ordinary readers, that may cause misunderstanding; for automated systems, it may translate directly into execution risk.
Mature probability infrastructure needs to provide both the price and its environment. The probability source, bid and ask quotes, market depth, trading volume, update timestamps, participant concentration, and resolution and settlement status together constitute the minimum context needed to interpret an implied probability.
Only when people can read this context, machines can access it, and the probabilities can undergo long-run calibration checks can market-implied probability develop from a prominent number into a reliable decision component. It does not promise to predict the future accurately; it offers a public judgment that is conditional, updatable, and open to evaluation against outcomes.
References
1. Charles F. Manski, “Interpreting the Predictions of Prediction Markets,” Economics Letters, 91(3), 425–429, 2006. DOI: 10.1016/j.econlet.2006.01.004.
2. Justin Wolfers and Eric Zitzewitz, “Interpreting Prediction Market Prices as Probabilities,” NBER Working Paper 12200, 2006. DOI: 10.3386/w12200.
3. Justin Wolfers and Eric Zitzewitz, “Prediction Markets,” Journal of Economic Perspectives, 18(2), 107–126, 2004. DOI: 10.1257/0895330041371321.