Prediction prices reveal more about whether than when
Whether Is Not When: Separating Occurrence from Timing in Polymarket Prices · arXiv:2610.36639
In plain words
A prediction market price can say an event is likely without saying it is close. This paper introduces a way to read contracts about the same event with different deadlines together, separating the chance the event happens from clues about when.
One day before their first qualifying Web report, prices crossed the study’s warning threshold for 56% of geopolitical events with a contract trading, compared with all 51 scheduled central-bank decisions. The study measures prices on one platform; it does not explain why traders set them.
The research
If most prediction markets settle No, accuracy can stay high while prices fail to warn about events that do happen. The question is whether market prices signal those events before they are reported, and whether a high price also tells us that an event is near.
The authors introduce a framework that reads contracts with different deadlines together, separating market information about whether an event happens from information about when.
To judge whether a prediction market gives an early warning, is it enough to see that an event is likely? The authors treat contracts on the same event with different deadlines as one forecast, then separate information about whether the event happens from information about its timing.
Across 12,573 ladder-days, prices carried more information about whether events happened than about when. In a separate seven-day warning test, the full ladder scored 0.30 in average precision, versus 0.25 for one contract; this score summarizes precision across recall levels. Reading deadlines together helped in this test, but did not make most warnings correct.
The event dates rely on first qualifying Web reports found by searches, which may have missed earlier reports. Some contracts opened too close to an event to measure earlier warnings, and the study is descriptive of geopolitical events on Polymarket.
What the evidence shows
- What prices warn about
The paper introduces a measurement framework for prediction market prices as early warnings. Its question is whether prices signal events that actually happen: because most Yes/No markets settle No, accuracy can look high even when prices miss events that happen.
The framework reads contracts on the same question with different deadlines together, so it can distinguish a signal that an event will occur from one about when it will occur.
- A threshold before the event
To compare warnings, the authors followed the highest-volume contract resolving Yes for each event and counted a price of at least 0.5 as a correct prediction. One day before the first qualifying Web report, the threshold was met for 109 of 193 geopolitical events with a trading contract, or 56%.
It was met for all 51 scheduled central-bank decisions one day before their official announcements. The event groups differ in type and size, so this is not a matched comparison.
- Whether outweighs timing
The framework then scores whether an event happens separately from when it happens. Across 12,573 ladder-days from 282 geopolitical-event ladders, prices added an average 0.25 bits per ladder-day about whether an event would happen beyond a base-rate reference.
They added 0.04 bits about timing beyond a reference that keeps the market’s occurrence chance but assumes the same daily chance until the last deadline. Timing was 13% of the information in prices under this score split.
- Deadlines improve seven-day warnings
The warning test asks whether reading all deadlines together helps predict an event within seven days. The authors compared the full-ladder estimate with one contract’s price over the same 12,573 ladder-days, including 494 followed by an event within seven days.
Average precision summarizes precision across recall levels; it is not the fraction of warnings that came true. The whole ladder scored 0.30, versus 0.25 for the single contract, supporting the value of deadline context in this test.
- What the findings establish
These results support a limited reading of market prices: they conveyed more information about whether an event would happen than about its date, while the full ladder gave better seven-day warnings than one contract. But a high price alone does not establish that an event is imminent.
Event dates depend on reports found by Web searches, some contracts opened too late to assess earlier warnings, and this descriptive analysis covers geopolitical events on Polymarket.
Deep
- Original paper authors
- Sizheng Fan, Yinghe Sun, Rong Zhao
- Submitted to arXiv
- License
- arXiv non-exclusive