About

Concept

Some prediction markets ask two questions at once. A combo market crosses a condition (does X happen?) with an outcome (does Y happen?). That gives four possibilities, each with its own price. Four prices like that are a full joint distribution, and a joint distribution gives you conditional odds by division:

P(Y | X) = P(X and Y) ÷ P(X)

Do the same division for "X does not happen" and you can compare the two: what traders expect for Y in a world with X, and in a world without it.

Example. One combo market crosses "Jensen Huang is out as Nvidia's CEO" with "Nvidia ends the year above a $4 trillion market cap". Dividing gives the odds Nvidia clears $4 trillion if he goes, and the odds if he stays.

Robin Hanson proposed markets like this in 1996, in Markets for Telling CEOs to Step Down, as a way to ask traders what a decision is worth before anyone makes it.

Estimation

Each page reads the live order books of the four combo outcomes, plus any ordinary markets that price a whole row or column of the same grid. In the example, those are a market on whether the CEO is out and a market on the company's market cap. That is more prices than unknowns, so they almost never agree exactly.

We look for the single 2×2 table that leaves the least arbitrage in those books. For any candidate table, every resting order priced on the wrong side of it is one an arbitrageur would take. The profit on those orders, added up over all the books, is that table's cost. The table with the lowest cost is our estimate.

This weights markets by the money in them. A quote backed by a few dollars is cheap to override. One backed by thousands is not.

Range

The range next to each number is every answer that costs within a set dollar amount of the best one: $10 unless you pick a larger amount with the Range buttons on a page. When the markets already agree, every answer that fits inside their bid-ask spreads costs nothing, so wide spreads give a wide range by themselves. The range tells you how much money stands behind the number. It is not a statistical confidence interval, and there is no probability attached to it.

The "Estimate depth" chart on each page draws the cost of every possible answer. A steep V means deep order books and a firm number. A shallow bowl means very little money would have to move for the answer to change.

Limitations

• It is not cause and effect. A world where the condition happens usually differs in other ways too. If a CEO is pushed out, there is probably other bad news, so lower odds "if out" would not mean removing them lowers the price.
• It is one line, not the whole picture. The outcome is a single yes-or-no question, such as a market cap above some level. You get the odds of clearing that line, not an expected value.
• The market's rules decide what counts. Read them on each market's page. In the example, "out" means a departure announced by the deadline, even if it takes effect later.
• Thin markets move easily. When little has traded, one small order can shift these numbers, and markets that disagree with each other are a sign nobody has arbitraged them into line. Each page shows how much has traded and how far apart its markets are.

Data

Prices and order books come from Polymarket's public API, read live by your browser. The history charts are recorded once a minute from the same order books. This site is not affiliated with Polymarket or with any company or person named in a market, and nothing here is investment advice.