Atlas Support
ResearchEconomicsWritten and edited by:

The Temporal Structure of Forecast Error on Polymarket: A Decomposition of Brier Loss over the 30 Days before Operational Market Closure

Seiryu Ando's SSRN working paper studies forecast error before operational market closure across 1,818 Polymarket markets and 1,393 events. This summary covers the August 15, 2026 revision.

Research question and measurement

How does the accuracy of market-implied probabilities change as a prediction market approaches closure? The paper measures Brier loss at six checkpoints during the preceding 30 days, using Gamma closedTime as the operational time anchor.

The analysis keeps the same cohort at every checkpoint, applies a 12-hour price-freshness requirement, and averages markets within events before weighting events equally.

Findings and interpretation

Mean Brier loss falls from 0.088253 at 30 days before closure to 0.034962 at 12 hours before closure, a decline of about 60.4%. In the Murphy decomposition, increased resolution accounts for the largest part of the improvement.

These results describe forecast errors in a fixed sample. They do not directly measure objective uncertainty or information arrival, and they do not establish continuous or monotonic acceleration in the improvement of forecasts.

Paper details and original text

An English-language working paper published on SSRN. The date shown for this Research entry is the revision date.

ItemDetails
Original titleThe Temporal Structure of Forecast Error on Polymarket: A Decomposition of Brier Loss over the 30 Days before Operational Market Closure
AuthorSeiryu Ando
Date written2026-07-21
First posted on SSRN2026-07-24
Last revised on SSRN2026-08-15
DOI10.2139/ssrn.7151878

Read the working paper

The full paper and its current version are available on SSRN.

Read on SSRN