Do prediction market prices mean what they say?
Explore 11,881 resolved Polymarket markets from the 1,500 closed events with the most trading, January 2024 to October 2026. Your browser recomputes everything below. It describes this sample, not prediction markets in general.
Testing the macro finding on new markets
In this sample, macro prices looked too timid (b = 1.66, or 1.29 after pooling). I found that after looking at the data, so markets that closed after the collection now test it, under rules frozen in advance.
How often Yes happened, against the market price
Markets grouped into ten price bins. Hover a point for its numbers.
Loading the market data…
Show as a table
| Price bin | Markets | Average price | Yes happened |
|---|
How to read it. A point above the dashed line means Yes happened more often than the price said. The slope b comes from logit(P(Yes)) = a + b·logit(price). Above 1, prices are too timid at the extremes: favourites are underpriced and long shots overpriced. Below 1, prices are too confident. The standard error is clustered by event, because markets in one event move together. An early close is a market that closed more than a day before its scheduled end, usually because the event already happened. A price measured back from that close already knows the answer. So the default view measures 7 days before the scheduled end, a time known in advance.
Reproduce every number
The study table is committed, so the report, charts and this page's data rebuild in seconds without the API.
git clone https://github.com/Guinerve/calibration-atlas cd calibration-atlas uv sync uv run calatlas report