Research

Live analysis of prediction market accuracy

Bell continuously measures how well Polymarket and Kalshi forecast elections and political events.

Live Last update:

0.08
Combined Brier Score
1 day before election
85.9%
Winner Prediction Accuracy
Across both platforms, shared elections
88.6%
Platform Correlation
Between shared elections
6,810
Electoral Markets
Election prediction markets

Aggregate statistics across all 15 political market categories, combining data from both Polymarket and Kalshi platforms.

Aggregate Statistics by Political Category

Market counts and trading volume across all political categories (Polymarket + Kalshi combined).

Electoral Markets by Election Type

Breakdown of electoral markets by election type, showing coverage on each platform.

We measure prediction accuracy using the Brier score—a metric where 0 is perfect and 1 is worst. Markets consistently achieve scores below 0.10, indicating strong predictive performance.

Accuracy Improves Near Election Day

Each line represents a cohort of markets open at least X days before resolution. Markets with longer trading windows (60d cohort) show lower Brier scores, indicating better predictions.

By Election Type

Brier scores across different election types (top 10 by volume).

By Political Category

Brier scores across all 15 political market categories.

Accuracy by Race Closeness

Brier scores by election margin. Close races (< 5% margin) are harder to predict, resulting in higher Brier scores.

A well-calibrated market means predictions match reality—when markets say 70%, outcomes should occur 70% of the time. We use quantile binning (equal sample sizes per bin) to ensure reliable estimates across the probability spectrum.

Predicted vs Actual Outcomes

When markets predict 70%, outcomes should occur 70% of the time. Points near the diagonal indicate accurate calibration.

Where Markets Place Their Bets

Markets favor confident predictions near 0% or 100%, with fewer predictions in the uncertain middle ground.

Market liquidity analysis based on historical orderbook data. Tighter spreads and deeper books indicate more liquid markets where traders can execute with less slippage.

Bid-Ask Spread by Category

Median relative spread (%) by political category. Lower spreads indicate tighter markets with better liquidity.

Liquidity vs Prediction Accuracy

Each line shows how Brier score changes across liquidity percentiles within a category. Downward slope = higher liquidity improves accuracy. Dot size = number of markets.

Spread Distribution by Platform

Distribution of relative spreads across platforms. Which platform offers tighter markets?

Spread vs Trading Volume

Do higher-volume markets have tighter spreads? Each point represents a market.

Liquidity Over Time

Daily median relative spread (%) across all active markets. Lower spreads indicate tighter, more liquid markets. Data available from Oct 2025.

Do prediction markets exhibit partisan bias? We analyze whether markets systematically over- or under-predict Republican victories, comparing calibration and regression results across both platforms.

Republican Win Probability Calibration

Predicted Republican win probability vs actual Republican win rate. Points above the diagonal indicate markets underestimate Republican chances; below indicates overestimation.

Partisan Bias Regression

OLS regression of prediction error on party and platform. Negative coefficients indicate the market was less confident in the winner (underpriced).

Distribution of Trader Partisanship

Among traders who bet for a party, what % of their volume was on that party? Broken down by trading activity. Polymarket only.

Distribution of Trader Accuracy

For each wallet, what % of their money was bet correctly? Broken down by trading activity level. One-time traders cluster at extremes (0% or 100%); active traders show more varied distributions. Polymarket only.

Trader Partisanship: Actual vs Perfect

What if all incorrect bets were flipped to the winning side? Compares actual partisanship with counterfactual. Filtered to traders with 2+ trades. Polymarket only.

Polymarket and Kalshi often list the same elections, allowing direct comparison. On shared elections, both platforms show remarkably similar predictions with 96.7% correlation.

Platform Agreement

Each point shows both platforms' final prediction for the winning candidate. Proximity to diagonal = agreement.

Head-to-Head Results

Which platform was more accurate on shared elections?

Platform Comparison

Key metrics across both prediction markets.

We track 23,000+ political prediction markets across both platforms, covering elections, policy, appointments, and more. Electoral markets represent the largest category by both count and volume.

Trading Volume Over Time

Monthly volume in millions USD. November 2024 saw record activity during the US presidential election.

Markets by Category

Distribution of political markets across 15 categories.

Prediction Confidence vs Trading Volume

How trading volume relates to prediction confidence. Points colored by outcome correctness. Use buttons to toggle platforms.