2008 — Obama v McCain
Obama won by 7.2pt
Markets: Markets called Obama strongly throughout
Polls: Polls trended Obama by 4-6pt on Election Day
Both correct on direction; markets had less day-to-day noise.
Compare · forecasting methodology
Prediction Markets vs Polls.
Five election cycles. Which won?
Two fundamentally different approaches to forecasting elections: real-money prediction markets that aggregate stake-weighted information continuously, and representative-sample polls measured episodically. Below: methodology comparison, the historical accuracy record across the 2008-2024 cycles, live market scale, and a frank assessment of when each method wins. Markets pegged Trump correctly in 2024; polls largely did not.
Active contracts
36,072
Kalshi contracts
28,772
Polymarket contracts
7,300
Refreshed
2026-06-08
Side by side
Methodology, incentives, scale.
| Prediction Markets | Polls | |
|---|---|---|
| Coverage | 36,072 active contracts (Kalshi + Polymarket) | Limited to funded questions |
| Methodology | Real-money contracts priced by traders | Representative sample surveys with statistical weighting |
| Incentive | Financial — wrong predictions lose real money | None for respondents; reputation only for pollsters |
| Update speed | Real-time (seconds) | Days to weeks per wave |
| Sample | Self-selected traders (skews male, educated, engaged) | Designed-representative (age, gender, geography, party-ID) |
| Cost to run | Near-zero marginal cost per question | $5K–$50K per high-quality poll |
| Manipulation risk | Expensive to manipulate (requires real money on the wrong side) | Herding, strategic responses, push polls, response weighting |
| Granularity | Binary outcomes; multi-outcome via separate contracts | Continuous percentages by demographic / issue |
| Long-tail coverage | Strong on liquid contracts; weak on niche topics | Weak — only well-funded topics polled |
| Why people answer | Profit motive | Phone-call/online cooperation, often <10% response rate in 2024 |
| Regulatory status | CFTC-regulated (Kalshi); offshore (Polymarket) | Unregulated; industry self-policing |
| Best for | Binary outcomes, real-time probabilities, head-line questions | Voter preference structure, issue salience, demographic breakdowns |
Historical record · 2008-2024
Five election cycles, head to head.
How prediction markets and polls each called the last five U.S. presidential cycles, with the actual outcome and a frank note on what each got right or wrong. Markets are not infallible — 2016 was a clear failure — but the directional record favors markets across most cycles.
2012 — Obama v Romney
Obama won by 4pt EC sweep
Markets: Markets pegged Obama 60-70% throughout October
Polls: Aggregators (538, RCP) showed near-tie at 50-50 in late Oct
Markets correct; polls under-weighted state-level Electoral College arithmetic. Nate Silver got it right; most polls did not.
2016 — Trump v Clinton
Trump won EC despite losing popular vote
Markets: Markets pegged Clinton 70-90% on Election Day
Polls: National polls showed Clinton +3-4pt; 538 gave Trump 28-29%
Both wrong on EC outcome; markets failed harder than 538's 28%. Polls had better state-level granularity but missed PA/MI/WI swing.
2020 — Biden v Trump
Biden won by ~4pt; close in 5 swing states
Markets: Markets called Biden 70-90% by Election Day
Polls: Polls overstated Biden margin (showed +8pt; result was +4pt)
Markets called direction correctly; polls had a +4pt systematic Trump under-count again.
2024 — Trump v Harris
Trump won swing states cleanly + popular vote
Markets: Polymarket pegged Trump 60-65%+ throughout final week; spiked to 70% Election Day
Polls: Polls showed near-tie; 538 gave Harris 50-50
Polymarket called direction correctly; polls had a third consecutive Trump under-count cycle. Markets visibly outperformed polls in 2024.
Live · current cycle
Markets pricing the current cycle right now.
Live snapshot of political contracts on Kalshi + Polymarket. Browse the full 36,072-contract universe at /odds or search natural-language at /ask.
FAQ
Common questions.
Are prediction markets more accurate than polls?
On most election cycles since 2008, yes — particularly for binary outcomes and final-week probability. Academic studies (Arrow et al., 2008; Wolfers + Zitzewitz; Rothschild, 2009) show prediction markets generally outperform polls on directional calls. The 2016 cycle was the major exception (markets failed harder than 538). 2024 was a clear market win — Polymarket pegged Trump 60-70% during the final week while polls showed a near-tie.
Why do prediction markets work?
Prediction markets aggregate private information from diverse participants who risk real money on their beliefs. This creates a strong incentive to be accurate rather than performative or socially-acceptable. The wisdom-of-crowds effect means that even if individual traders are biased, the aggregate price (set by trades from informed traders willing to bet against the consensus) tends toward the true probability. Polls do not have this incentive structure — respondents face no cost for inaccurate or strategic answers.
When should I trust polls over markets?
Trust polls for understanding WHY people hold certain views — demographics, issue rankings, approval breakdowns by sub-population, longitudinal opinion shifts. Polls answer "what do people think and how does it differ across groups". Trust markets for the probability of a specific binary outcome. The two are complementary: pollsters tell you the structure of opinion; markets tell you the probability of a result. Sophisticated forecasters consume both.
Can prediction markets be manipulated?
Yes, but expensively. To move a liquid Polymarket or Kalshi contract you need real USDC or USD on the wrong side; sophisticated traders will fade you, eating your capital. Studies of attempted manipulation (e.g. 2012 InTrade Romney pump) show effects last hours, not days. Polls, by contrast, can be moved by small sample biases, push-poll wording, or strategic non-response — at near-zero cost to the manipulator.
Why did prediction markets miss 2016 if they're so good?
Markets are not omniscient — they reflect the best aggregate of available information. In 2016 the same systematic state-level polling errors that fooled 538's model (PA, MI, WI under-counted Trump support) also fed into market prices, since traders read polls as their primary input. The market failure in 2016 was downstream of the polling failure, not independent of it. By 2020 and 2024 markets weighted state-level dynamics more heavily and corrected.
How do I track prediction market accuracy in real time?
SimpleFunctions publishes live Brier calibration scores at /api/calibration — currently Kalshi 0.20 and Polymarket 0.12 on T-24h price over the past 90 days. The /calibration page breaks scores down by category, time-to-resolution, and price bucket so you can audit accuracy on the question class you care about. Polls have no equivalent live calibration endpoint — accuracy is typically only assessed retrospectively after the cycle.
Can I bet on poll outcomes themselves?
Yes — both Kalshi and Polymarket list contracts on poll-related outcomes (e.g. "Will Trump's 538 average exceed X% on date Y?"). These are second-order markets-on-polls and are a meta-forecasting layer. They tend to track polling-aggregator releases tightly. SimpleFunctions surfaces these in /api/public/markets like any other contract.
What about academic research on this question?
Foundational papers: Arrow, Forsythe, Gorham, Hahn et al. "The Promise of Prediction Markets" (Science, 2008) argues for legalizing markets for forecasting. Wolfers + Zitzewitz "Prediction Markets" (J. Econ. Perspectives, 2004) reviews the evidence. Rothschild "Forecasting Elections: Comparing Prediction Markets, Polls, and Their Biases" (POQ 2009) is the canonical empirical comparison. The general finding: markets win on directional accuracy on liquid binary questions; polls win on richer demographic and issue information.
Related
Deeper guides + neighboring comparisons.
- How to backtest a prediction market strategy
Use settlement data to build calibration curves and measure realized returns.
- Edge calculation in prediction markets — theory to execution
Theoretical edge vs depth-adjusted edge, and when to walk away from a trade.
- Kalshi vs Polymarket — flagship head-to-head
Live cross-venue spreads, Brier calibration, fee math.
- Polymarket vs Metaculus — real money vs community
When the panel-forecasting layer wins.
- Live calibration leaderboard
SimpleFunctions\'s own Brier scores per venue, category, and price bucket.
Live odds, no polls.
SimpleFunctions normalizes Kalshi + Polymarket into one schema with live prices, calibration scores, and indicator screening. Public reads free, no auth required.