Polymarket's accuracy in past elections is generally strong—especially close to resolution and on high-volume national races—but mixed overall, with notable strengths over polls in 2024 and limitations in efficiency, lower-volume markets, and earlier cycles.

Key Metrics for Evaluation

• Accuracy rate: Percentage of markets resolved correctly (e.g., favorite wins).
• Brier score: Proper scoring rule for probabilistic forecasts (lower is better; 0 = perfect, 1 = worst; ~0.25 for random). Sports oddsmakers often land ~0.18–0.22.
• Calibration: How well implied probabilities match real outcomes (e.g., events priced at 70% should happen ~70% of the time).
• Comparison: Vs. polls, models (e.g., FiveThirtyEight, Nate Silver), or other markets.

Polymarket (launched 2020) has limited pre-2024 election history, with explosive growth in 2024 politics ($3+ billion volume on the presidential race alone).

2024 U.S. Presidential Election (Strong Performance)

Polymarket was one of the better forecasters:

• Correctly signaled Trump's victory when many polls/models showed a toss-up or slight Harris edge. Late markets had Trump as a clear favorite (~55-60%+ in final stretches).
• Outperformed national polls in swing states per Bayesian analyses (Vanderbilt study). Implied probabilities tracked closely to final vote shares in key areas.
• Early calls: Predicted Biden withdrawal probability spiking to 70% post-debate (well before announcement). Correctly favored Walz over Shapiro as VP pick.
• Caveats: Large "whale" bets (e.g., one French trader ~$30M+ on Trump) caused swings and potential temporary distortions. Diverged from other markets/polls mid-cycle. One analysis of thousands of 2024 political markets found Polymarket at ~67% accuracy (lower than PredictIt ~93%, Kalshi ~78%), with inefficiencies like arbitrage opportunities.

Overall, a win for the platform and "wisdom of crowds" narrative in a high-liquidity environment.

Broader Track Record (2020s Elections and Markets)

• Political markets: Strongest category. One analysis (thousands of resolved markets): ~81% accuracy on binary electoral outcomes, ~74% multi-candidate, Brier ~0.164. Overall platform Brier often ~0.047–0.187 depending on scope/timeframe.
• Near-event accuracy: ~90% a month out, up to 94% hours before resolution (per Dune dashboard analyses excluding extremes). Politics performs well close to election day.
• Pre-2024/Other cycles: Limited U.S. presidential data (platform young). Mixed results in some international or lower-volume races; markets can lag local info or be overconfident early. Some studies show prediction markets historically competitive with or better than polls but not infallible (e.g., past cycles had misses like over-favoring certain candidates).
• Midterms/Other: Competitive with models; accurate on clear favorites but noisier on close/niche races (e.g., congressional). Lower liquidity leads to more variance.

Strengths

• Financial incentives + crowd aggregation often incorporate info faster than polls (e.g., scandals, turnout signals).
• High volume improves reliability (presidential >> down-ballot).
• Transparent, real-time, and probabilistic.

Weaknesses and Criticisms

• Manipulation/Whales: Large bets can skew prices temporarily (investigated but often not "manipulative" in intent).
• Inefficiencies: Prices diverge across platforms; not always efficient (arbitrage exists); can reflect trader demographics more than electorate.
• Not a crystal ball: Overconfident early; weaker on low-volume or complex outcomes. One view: Good single data point in 2024 but jury still out broadly.
• Bias risks: Crypto-heavy user base may lean certain ways; volume concentration matters.

In summary, Polymarket has built a credible track record, particularly shining in the high-stakes 2024 cycle where it outperformed polls on the presidential outcome. It's a useful signal (especially late and on major races) but best used alongside polls, fundamentals, and other data—not in isolation. For gubernatorial or state races like Kansas 2026, expect solid directional accuracy if volume builds, but watch for liquidity issues. Performance improves as events near.

Sources include academic analyses, Dune dashboards, platform data, and post-election reviews. Accuracy claims can vary by methodology/time window.