Is Kalshi or Polymarket more accurate?

On the 89 decided 2024–25 races both exchanges listed, Polymarket's eve-of-election prices scored a Brier score of 0.0645 and Kalshi's 0.066 (lower is better).

What happens to races the markets call close?

Favorites priced between 50% and 60% won 50% of the time in 2024–25, about what their prices implied: close races really are close.

Track record

How often the markets get it right

Across 125 decided 2024 and 2025 races, the prediction-market favorite on the eve of the election won 111 (89%). Favorites priced at 80% or more won 86 of 86. Races priced near a coin flip (50 to 60%) went to the favorite 50% of the time.

Do favorites win as often as their price says?

Each row groups races by the favorite's price, and compares that price with how often the favorite actually won. A well-calibrated market's two bars match.

The day before 111 of 125 favorites won · Brier 0.0722

50–60%16 racespriced 55%won 50%

60–70%11 racespriced 64%won 82%

70–80%12 racespriced 75%won 67%

80–90%11 racespriced 85%won 100%

90–100%75 racespriced 97%won 100%

A week before 106 of 111 favorites won · Brier 0.0442

50–60%7 racespriced 54%won 57%

60–70%14 racespriced 64%won 86%

70–80%9 racespriced 74%won 100%

80–90%6 racespriced 85%won 100%

90–100%75 racespriced 97%won 100%

A month before 98 of 107 favorites won · Brier 0.053

50–60%13 racespriced 54%won 46%

60–70%7 racespriced 65%won 86%

70–80%5 racespriced 77%won 100%

80–90%11 racespriced 86%won 91%

90–100%71 racespriced 98%won 100%

Price = the average of Kalshi's and Polymarket's latest readings in the week up to each date, as the Democrat's share of the two main parties. Brier score: the average squared miss, 0 is perfect, 0.25 is a coin flip.

Kalshi vs. Polymarket

Exchange	Races listed	Favorite won	Brier, all	Brier, shared races
Kalshi	104	92 (88%)	0.0829	0.066
Polymarket	110	101 (92%)	0.0567	0.0645

"Shared races" are the 89 races both exchanges listed, the fair head-to-head.

The Parlay estimate

Our model nudges clear favorites further toward winning, because in these races they won more often than priced. Tested on races it wasn't fit to, it scored a Brier of 0.07406 against the market's 0.0756. How the model works →

The upsets: 14 favorites that lost

2024 NE-02 House: Tony Vargas was favored at 75%; Don Bacon won.
2024 PA Senate: Bob Casey Jr. was favored at 75%; David McCormick won.
2024 PA-07 House: Susan Wild was favored at 72%; Ryan Mackenzie won.
2024 PA-08 House: Matt Cartwright was favored at 71%; Rob Bresnahan Jr. won.
2024 CO-08 House: Yadira Caraveo was favored at 66%; Gabe Evans won.
2024 MI presidential: Kamala Harris was favored at 63%; Donald Trump won.
2024 IA-01 House: Christina Bohannan was favored at 59%; Mariannette Miller-Meeks won.
2024 WI presidential: Kamala Harris was favored at 58%; Donald Trump won.
2024 OR-05 House: Lori Chavez-DeRemer was favored at 56%; Janelle Bynum won.
2024 IA-03 House: Lanon Baccam was favored at 56%; Zach Nunn won.
2024 CA-45 House: Michelle Steel was favored at 53%; Derek Tran won.
2024 AK-AL House: Mary Peltola was favored at 53%; Nicholas Begich III won.
2024 CA-22 House: Rudy Salas was favored at 52%; David G. Valadao won.
2024 PA-10 House: Janelle Stelson was favored at 51%; Scott Perry won.

2026

We've logged 5,510 hourly readings of 551 races. Every 2026 race is scored here as it settles after Election Day, November 3, using the odds recorded before the result.

Decided 2024 and 2025 U.S. races on Kalshi and Polymarket (ingest/model/training.py). Download the race-by-race data: track-record.json. Research, not betting advice.