HIP-4 Research
Does the market know what it is talking about?
A prediction market states a probability. This page checks it. For every settled HIP-4 market we take the price traders actually paid for a side, bucket it, and count how often that side won.
Backing the favouriteEvery buy at 50¢ or more on a crypto price market: how much more often that side won than its price implied, in points on a $1 contract. It is not a return on money: a side bought at 85¢ risks 85¢ to make 15¢. The price family leads because no single market owns more than about 7% of any of its bands.
+3.3 pts
per $1 contract, crypto price markets, 16,898 markets
Trades studied
6,420,476
from 18,832 settled markets
Staked on those favouritesThe dollars that actually went through the book on those trades, not the archive's whole volume, which also counts the cheap side of every market.
$93.18M
+2.8% per dollar staked
What the price implied, and what happenedTrades grouped by the price paid, both sides of every market; the last column is how often that side won.
Backing the favourite: +1.2 pts per $1 contract (6,420,476 trades from 18,832 markets)
- 0–10¢285,7773.2%1.7%
- 10–20¢388,84715.7%18.8%
- 20–30¢769,28724.9%21.9%
- 30–40¢779,98935.3%30.3%
- 40–50¢995,36545.0%45.5%
- 50–60¢949,94054.9%56.3%
- 60–70¢778,83264.4%65.6%
- 70–80¢569,73875.0%80.9%
- 80–90¢489,78384.6%79.5%
- 90–100¢412,91896.0%97.6%
Green: the side won more often than its price implied. Red: less often. Within two points either way, neither.
Read the bands with this in mindHow much of a price band one single market accounts for.
- Sports & events: in the 80-90¢ band, 2026 World Cup Champion: Spain alone accounts for 21.9% of the 264,445 trades there, and those trades were on the side that lost.
- The fills of one market settle together, so a band one market dominates is close to a single observation however many trades it holds. That is why both counts are printed everywhere, and why the per-market view is kept beside the per-trade one.
- The switch above the table draws the same study without them. It removes every market holding 15% or more of a band, 1 in this archive, and it is a check rather than the answer: the figure this page states is the one measured over everything.
Method, and what it does not say
- The unit is the trade. Every buy of a settled market is a claim at the price it paid, on the side it bought, and it won if that side won. Trades are not independent: the fills of one market settle together, so the number of markets behind them is printed beside them, and the per-market view (one vote per market) is kept as the check.
- The headline is an edge in points on a $1 contract, not a return on money. A side bought at 85¢ pays 15¢ on 85¢ risked, so the same edge is a larger return the dearer the favourite is. The two are one division apart and are never printed as the same number.
- One dollar on each of those favourite trades returned +4.4% on average, and the dollars actually staked returned +2.8%. Both are measured over the same 16,898 markets as the curve: a return is published only once every one of them carries the size of each trade, because the archive is re-read newest first and a half-read archive is a biased slice of it rather than a small one.
- The same measure with one vote per market instead of one per trade gives +7.6 pts over 18,832 settled markets. It is kept as a check rather than drawn as a curve: it weighs a $150 market like a $20M one, and files a market that walked from 30¢ to 90¢ at 60¢, a price nobody ever paid.
- Every buy of a settled market is counted once, at the price it paid and on the side it bought. Both legs share one histogram, so a No bought at 30¢ is a trade at 30¢ like any other. Mints, merges and settlements print at 0.50, 1 or 0 (mechanics, not opinions) and are left out.
- A buy is graded as held to settlement. Nobody's exit is followed, so this measures the price and not what traders earned: someone who bought at 30¢ and sold at 60¢ before the end counts here as a 30¢ buy that lost, if that side lost. Selling is not invisible either. On Hyperliquid it books as a buy of the other side at 1 minus the price, and is counted in its own right.
- It is an average over the market's whole life, not the closing line. It measures what traders paid, not what the book last thought.
- Markets whose Yes leg never traded carry no price and are excluded, as are draws: a two-sided match that ends level pays both sides $0.50 and has no winning leg.
- The two families are folded separately on purpose. They share no traders, no deployers and no resolution source, so a bias present in both is a property of the market rather than of one crowd.