Prediction Markets, Explained

The easiest way to understand a prediction market is to forget the financial terminology for a moment and start with a question: will something happen, or will it not? A market lets people buy and sell positions linked to the possible answers, and the price moves as participants change their minds or new information arrives.

If a YES contract that pays one dollar when an event happens is trading around 63 cents, people often describe that as roughly a 63% implied probability. I would not call it a scientific forecast. It is better understood as the current market price attached to one very specific question, under one very specific set of rules.

It means the market is currently pricing the YES side at about 63 cents on the dollar, not that somebody has proved the event has a 63% chance of happening. A weather forecast is a useful comparison: a 70% chance of rain does not become nonsense because the day happens to stay dry. Probability describes uncertainty; it does not remove it.

That distinction matters because prediction-market numbers can look more precise than the underlying question really is. A number with two decimal places is still only useful if we understand what it measures.

Because the largest number on the screen is often the least useful place to begin. Before I care whether a market says 42% or 78%, I want to know what exactly has to happen, by when, and according to whose rules.

Building Prediction Market Radar reinforced this repeatedly. Two markets can look almost identical at headline level and still resolve differently because of a deadline, a jurisdiction, an eligibility rule, a threshold or a different settlement source.

Similar words are not the same thing as the same outcome. A football market asking whether a team will win the match is not necessarily equivalent to one asking whether the team will qualify for the next round. Extra time, penalties and competition rules can make those two questions resolve differently.

From a Computing point of view, this is an identity problem. If two databases both contain a row called “Apple”, I cannot safely join them until I know whether each row means Apple Inc., the fruit, a record label or something else. Prediction markets have the same problem at much larger scale.

Different platforms have different participants, liquidity, fees, access rules and information flows. Even if two contracts are materially compatible, their prices can diverge.

I sometimes think of it as asking two different rooms of people to estimate how many sweets are in a jar. There is one true answer, but the two groups may arrive at different estimates because they contain different people and different information. The disagreement may be interesting — but first I want to make sure both rooms are looking at the same jar.

A Super Match is Prediction Market Radar’s way of saying that markets from different sources appear sufficiently compatible to compare. I deliberately prefer “compatible” to pretending every source contract is word-for-word identical.

The source wording and links remain visible so the reader can inspect the evidence. If the system is not confident enough, the preference is to withhold the comparison rather than manufacture certainty just to increase the number of matches.

Settlement is where the question becomes real. “Will Candidate A win?” sounds simple until you ask which election, which jurisdiction, what counts as winning, what happens if the candidate withdraws, and which official source determines the result.

A prediction-market comparison that ignores settlement wording can be beautifully formatted and still be wrong. That is why settlement compatibility is part of the identity check rather than a footnote.

A market is the question or proposition. An outcome is one of the possible ways that question can resolve. In a binary market, the outcomes are normally YES and NO. In a multi-outcome election market, the outcomes might be several candidates.

This distinction matters across platforms because one venue may represent an event as one multi-outcome market while another splits the same event into several binary contracts. The words can look different even when the underlying economic question is closely related.

That was a deliberate design choice. I wanted to solve the reference-data problem first: what does each market mean, where did it come from, when was it observed, and which other markets are genuinely comparable?

Prediction Market Radar does not need to hold funds, connect a wallet or execute a trade to be useful. Keeping the comparison layer independent and read-only makes the purpose clearer: evidence, identity, provenance and comparison.

Because a missing match is better than a false one. If the evidence is uncertain, saying “I am not confident enough” is useful information.

Coverage is important, but trust is more important. A system that always produces an answer can be less useful than one that knows when not to.

Prediction Market Radar refreshes from its supported sources on an automated cycle. The important point is that freshness is visible rather than implied. A market price without a timestamp can be misleading, particularly around elections, sporting events, regulatory announcements or central-bank decisions.

I would rather show an older timestamp honestly than make the interface look permanently “live” when the source itself has not changed.

Consider “Will the Federal Reserve cut rates at the September meeting?” and “Will the target rate be lower by the end of September?” They sound almost interchangeable, but one is tied to a meeting and the other to a date. An unexpected action outside the meeting could make the answers diverge.

The same problem appears in election filing deadlines, sports competition stages, product launch windows, award eligibility periods and economic-data release dates. Time is part of identity.

Of course. A 90% price does not mean the event is guaranteed. If we observed a large number of genuinely comparable situations priced around 90%, we would expect some of them not to happen.

This is why I avoid presenting prediction markets as crystal balls. They are useful information systems for uncertainty, and part of their value is how quickly the collective estimate changes when new evidence appears.

No. A poll asks selected people what they think or intend to do. A prediction market produces a price through trading activity. Those are different mechanisms with different strengths and biases.

For an election, I would not treat a poll and a market as substitutes. They are different evidence streams. The interesting question is often why they agree or disagree.

Because disagreement can tell us something — but only after identity has been established. If two materially compatible markets show meaningfully different prices, that may reflect liquidity, participants, information flow or market structure.

The useful product is not merely a table of percentages. It is a defensible bridge between the underlying contracts so that the comparison itself has meaning.

Because I want the comparison to be auditable. A reader should be able to leave Prediction Market Radar, inspect the original contract and decide whether they agree with the comparison.

If a system asks people to trust its interpretation while hiding the underlying evidence, it becomes very difficult to distinguish confidence from convenience.

No. It is a comparison and research layer. I am much more interested in helping someone understand what a number belongs to than telling them what action to take because of that number.

That distinction keeps the product useful to people who are not traders at all — journalists, researchers, platforms, analysts and anyone simply trying to understand how different prediction markets describe the same event.