PDATA Analyses 39,681 Markets to Test Polymarket and Kalshi Accuracy

United States.- 8 October 2026 | www.zonadeazar.com PDATA has published a new series of prediction-market research reports using data collected across eight platforms, with one study focusing on a central industry question: how closely market prices reflect actual probabilities.

Its accuracy analysis covered 39,681 binary markets from Polymarket and Kalshi that settled between 28 September and 5 October 2026.

Average Gap of Just One Point

The study compared prices recorded 24 hours before market close with eventual outcomes.

When grouped into 10-point probability bands, prices were on average only 1 percentage point away from the actual share of outcomes that occurred.

Favourites Won 83% of the Time

The side priced above 50% ultimately won 83% of the time.

PDATA said this indicates strong overall calibration, while also identifying several small but systematic differences between market prices and realised outcomes.

Favourite-Longshot Bias

Low-probability outcomes occurred slightly less often than their prices implied.

Higher-probability outcomes in the 70% to 90% range, meanwhile, won somewhat more often than expected.

The pattern resembles the favourite-longshot bias commonly identified in forecasting and betting markets.

Markets Below 10%

Low-probability contracts made up a large part of the sample.

A total of 16,644 markets, or 42% of the dataset, were priced below 10%.

Their average price was 2.2%, while only 1.4% ultimately resolved YES.

Polymarket and Kalshi Separately

Both platforms showed similar overall calibration.

Polymarket’s average calibration gap was 0.9 percentage points, compared with 1.1 points for Kalshi.

PDATA cautioned against directly comparing their Brier Scores because the two platforms list materially different types of markets.

Sports Dominate the Dataset

Sports represented 26,484 of the 39,681 markets analysed, accounting for approximately 67% of the total sample.

The category produced an average calibration gap of around 1 percentage point.

Elections and Politics

Election-related markets also produced relatively small calibration gaps.

The Elections category recorded an average gap of 1.3 points, while Politics came in at 2.1 points.

PDATA stressed that the findings cover only the period studied and should not automatically be extended to full election cycles or longer-duration political markets.

Seven Days Before Close

The research also examined prices recorded seven days before markets closed.

That sample contained 6,469 markets and produced an average calibration gap of around 1.4 percentage points.

The favourite won 94% of the time, although the sample was heavily concentrated in highly lopsided markets.

Study Limitations

PDATA emphasised that the report covers only eight days of settled markets.

The dataset is also heavily weighted towards sport, meaning the findings should not automatically be applied to elections, finance or other long-dated events.

The study also excludes markets without meaningful trading activity or with bid-ask spreads greater than 10 points, avoiding the treatment of inactive quotes as genuine forecasts.

Eight Platforms in One Dataset

PDATA collects prices, volume and outcomes across:

  • Polymarket.
  • Kalshi.
  • Manifold.
  • Myriad.
  • Limitless.
  • Predict.
  • Opinion.
  • Gemini.

Data is refreshed at intervals ranging from 10 to 60 minutes depending on the venue.

Wider Research Programme

Alongside the accuracy study, PDATA has published reports examining:

  • How each prediction market defines volume.
  • What data each platform API provides.
  • Price differences for the same question across venues.
  • Market distribution by category.
  • Cross-platform data comparability.

Open Data Access

PDATA makes the underlying data publicly available under a CC BY 4.0 licence.

Users can access it through:

  • API.
  • Nightly bulk files.
  • Public read-only Postgres.
  • MCP server for AI assistants.

Next Steps or Impact

The study provides an early quantitative measure of how closely Polymarket and Kalshi prices behave like calibrated probabilities.

For operators, regulators, traders and researchers, comparable cross-platform data can help distinguish headline volume and liquidity from actual market accuracy.

PDATA plans to continue updating the research as additional settlement windows and market data become available.

Editó: @fonta

Compartir: