Prediction markets: how they actually work
A prediction market turns a question into a tradable contract. "Will X happen by date Y" becomes a claim that pays one dollar if it does and nothing if it does not, and that claim trades on an order book at some price between the two.
That much is easy to explain, and most introductions stop there. The mechanics underneath are where the money is made and lost — and they differ enough between venues that "prediction market" describes a shape rather than a single thing. This page is an index of those mechanics, with the detail in the linked articles.
Price, and its relationship to probability
The headline claim about prediction markets is that prices are probabilities. It is close enough to be useful and wrong enough to be worth understanding.
A price of 0.30 on a contract that pays a dollar means the last trader was willing to value that outcome at thirty cents. It is not the output of a model, nobody is enforcing calibration, and the number can be stale, thin, or simply mistaken. The strongest version of the claim is that prices aggregate belief, weighted by willingness to stake money — which is genuinely better than a poll, and not the same as a probability.
The clearest demonstration is what happens across a set of outcomes where exactly one can win. The prices should sum to one. Very often they do not, because each outcome trades in its own book against its own buyers and sellers, and nothing mechanically ties them together.
Read next: Why don't prediction market odds add up to 100%? — where the gap comes from, why it persists, and what it actually costs to close, which is considerably more than the gap appears to be worth.
What you own when you buy an outcome
This is the question that decides everything else, and the answer is venue-dependent.
On some venues, buying an outcome mints you an on-chain token representing that claim. You hold it in a wallet, and when the market resolves you redeem it for the payout — a transaction, which you or your platform submits. On others, the outcome is a market on an exchange order book, and buying opens a position in your exchange account that the venue settles for you at resolution.
Both let you sell before resolution. Both pay out if you are right. But the failure modes are different, the collateral behaves differently, and the operational risks are not the same ones.
Read next: How are prediction markets on Hyperliquid different from Polymarket? — the token-versus-position distinction and everything that follows from it, including order price rules and minimum sizes.
Getting money out
Buying is instant. Getting paid frequently is not, and the reasons have little to do with block times.
Before a platform can move your balance it has to establish that the balance has stopped changing, because settlement and redemption pay out asynchronously and a snapshot taken mid-flight moves the wrong amount. Afterwards it has to establish that the funds actually arrived — which sounds trivial and is the step most likely to be done wrongly, because a destination balance going up by roughly the right amount is not the same as your withdrawal having landed.
There is also a floor below which withdrawing destroys value, since network fees can exceed a small balance outright.
Read next: Why does withdrawing from a prediction market take minutes, not seconds? — the settling wait, the dust threshold, and why a matching balance change is not proof of delivery.
When a market ends badly
Most markets resolve cleanly. The interesting engineering is in the ones that do not.
A market can be removed before it resolves — cancelled, voided, or migrated — leaving a position whose value nobody has established. The tempting fix is to settle it at the last traded price, which is wrong in a specific and expensive way: a contract that pays zero or one is never actually worth the forty cents it last changed hands at, so settling there pays out an amount that was never a possible outcome.
Getting this right means deciding what evidence is sufficient to establish an outcome, and being willing to leave a position visibly unresolved when none is.
Read next: What happens to your position if a market is delisted before it resolves? — the two independent evidence sources, why they must agree, and why guessing is never the fallback.
If you are trading these
A short version of what the articles above add up to.
Read the book, not the headline price. The displayed odds tell you where the last trade happened, not what you can get filled at. On thin markets those differ a lot, and the difference is exactly what eats an apparent arbitrage.
Know which venue model you are on. Whether you hold a redeemable token or an exchange position determines whether your risk is a failed claim transaction or a withdrawn listing — and, on an exchange account, whether your outcome position shares collateral with everything else you have open.
Assume exits are slower than entries. Not because platforms are slow, but because paying out correctly requires establishing things that buying does not.
Test with a real amount, not a trivial one. Dust thresholds mean a two-dollar test can behave completely differently from a two-hundred-dollar one, and discovering that on the way out is a poor time to learn it.
BT365 runs prediction markets on both Polymarket and Hyperliquid, alongside perps and spot, from one self-custodial account. Explore the platform or read the documentation.