The world’s first prediction market for private, unlisted companies.

NOTSTOCKS
Sign in
Back to blog

When to Trust Prediction Market Accuracy, Practical Rules and NotStocks

Prediction market accuracy title card illustration

Prediction markets are generally informative and often outperform polls in short- and long-horizon forecasting, but that accuracy is conditional on domain, time-to-resolution, and liquidity. Research from the Iowa Electronic Markets shows markets beating polls across election cycles, while newer studies flag systematic miscalibration in thin or long-dated markets. The rest of this piece unpacks how researchers measure that accuracy and when a market price deserves your trust.

TL;DR:

  • Prediction markets tend to be more accurate than polls in election forecasting, especially when contracts are liquid and close to resolution.
  • Calibration declines as the time-to-resolution increases and in thin or long-dated political markets, which often understate genuine confidence.
  • Proper scoring rules, like the Brier score, ensure forecasts reflect honest beliefs, but market prices can drift due to biases like the favorite-longshot effect or cross-market disparities.
  • Most profit in prediction markets goes to a small, disciplined group of traders who provide liquidity, rather than the average participant.
  • The reliability of a market price depends on liquidity, recency, and domain, making it a useful signal only under specific conditions.
Notstocks
Explore Signals From Private Markets
NotStocks lets you trade on the future success of private companies across healthtech, fintech, AI, and other emerging sectors.
Explore NotStocks

Table of Contents

What the evidence says about prediction market accuracy

The strongest long-run case for prediction markets comes from election studies. Analysis of the Iowa Electronic Markets found that market prices tracked eventual outcomes more closely than contemporaneous polls, both in the final days before an election and months out, and the researchers built methods to estimate forecast standard errors directly from price time series rather than relying on poll-style margins of error.

More recent cross-platform work complicates the picture. A large analysis of trades on Kalshi and Polymarket found that calibration and accuracy vary systematically by domain, horizon, and trade size rather than holding constant across a platform, and political contracts in particular tend to compress toward 50%, understating genuine confidence.

A few patterns show up consistently across this research:

  • Markets tend to beat unadjusted polls on election forecasting over both short and long horizons.
  • Accuracy is not uniform: it shifts with how close the market is to resolution and how liquid the contract is.
  • Political markets show more compression toward the midpoint than markets in other domains.

Across a dataset spanning 353 million trades on 429,000 binary contracts, calibration quality depended heavily on domain and horizon rather than the platform alone. That scale of evidence is why blanket claims like “markets are always accurate” or “markets are always wrong” both miss the point.

How researchers actually measure forecast accuracy

Most academic accuracy claims rest on the Brier score, which measures the squared difference between a predicted probability and the actual outcome (0 or 1). A forecast of 0.90 for an event that happens scores 0.01, close to perfect, while the same forecast on an event that does not happen scores 0.81. Lower scores mean better calibration, and the score rewards confident, correct predictions far more than hedged ones.

The Brier score belongs to a family of proper scoring rules, including the log score, which are constructed so that a forecaster’s best strategy is to report their true belief rather than to game the metric. Market scoring rules, such as the logarithmic market scoring rule, extend this logic into a trading mechanism: the market itself pays out based on a proper scoring rule, which is part of why prices can function as calibrated probability estimates rather than arbitrary betting odds.

  • The Brier score penalizes overconfidence and rewards accuracy, not just direction.
  • Proper scoring rules make honesty the dominant strategy for anyone submitting a forecast.
  • Market scoring rules like LMSR translate that same incentive into continuous trading.

Unlike a poll, a market has no simple sample-size formula for its margin of error. Researchers instead estimate forecast standard errors from the market’s own price history and from comparisons across related contracts, treating the price path itself as the data.

Pro Tip: When comparing two forecasts, ask which one was scored with a proper scoring rule before trusting a claim that one method “beat” another.

Why market prices sometimes miscalibrate

Prediction market prices drift away from true probabilities for identifiable reasons, and knowing them helps you judge any specific price you see.

  1. Time-to-resolution effects. Calibration tends to worsen the further a contract sits from its resolution date, and research on this longshot bias found a persistent S-shaped pattern where a price around 0.20 corresponded to a realized frequency closer to 15.3% in some samples.
  2. Domain differences. Political contracts often compress toward 50%, understating real confidence, while sports and weather markets, which resolve quickly and repeatedly, tend to show tighter calibration.
  3. Liquidity and trade size. Thin markets move more on a single large trade, and cross-platform analysis found that trade size itself explains a meaningful share of calibration variance.
  4. Favourite-longshot bias. Low-priced, longshot contracts are consistently overvalued relative to their true odds, a pattern that shows up across many prediction market datasets and betting markets generally.
  5. Cross-market divergence. The same event can price differently on different platforms due to differences in trader populations, fee structures, and available liquidity, which means no single platform’s price should be treated as the final word.

Who actually profits from prediction markets

Profits in prediction markets are not distributed evenly across participants. Practitioner-focused research on trader returns has found that a small share of traders, often those providing continuous liquidity through limit orders, captures most of the gains, while the majority of participants roughly break even or lose money over time.

  • Being a skilled forecaster and being a profitable trader are related but distinct skills; one requires good judgment, the other requires disciplined execution and patience with order placement.
  • Persistent limit orders that capture the bid-ask spread tend to generate steadier returns than repeated market orders based on a hunch.
  • Individuals entering a market for the first time should expect modest edges at best, not consistent windfalls.

Pro Tip: Scale your position to the actual gap between your estimate and the market price rather than betting a flat amount on every contract you feel strongly about.

When to trust a market price and when to be careful

A market price is worth more weight when several conditions line up at once, and less when they do not.

  • Favor contracts with a short time-to-resolution, since calibration tends to sharpen as the event approaches.
  • Look for active trading volume and a narrow bid-ask spread, both signs that the price reflects real information rather than a stale quote.
  • Check whether the domain has a track record of good calibration; political contracts warrant more skepticism than frequently repeated sports or weather bets.
  • Treat long-horizon political prices as directionally useful but likely underconfident, and adjust your own estimate accordingly rather than taking the number at face value.

Combining market prices with structured self-reports can outperform either signal alone: aggregated, properly weighted belief estimates have matched or exceeded market prices in controlled forecasting tournaments. Treat a single price as one input, not a verdict.

Where NotStocks fits into this accuracy picture

Private, unlisted companies present a sharper version of the accuracy problems described above: information is scarce, trading is thinner, and there is no public price to anchor expectations. NotStocks is built around that gap, offering markets on private companies across sectors like healthtech, fintech, and AI, with a visible price history and daily analysis for each listing, such as its Outbid.lol market page. That transparency does not eliminate the calibration challenges liquidity and horizon create, but it gives traders more of the context they need to judge a price on its merits rather than trading blind.

Private market forecast factors illustrated

What matters most in this debate

What matters most in this debate — overview diagram

The loudest claims about prediction markets tend to be the least useful ones: either “markets are always right” or “markets are just gambling.” Neither survives contact with the evidence. What actually matters is horizon and liquidity, not the existence of a market itself. A liquid, short-dated contract in a domain with a track record of good calibration is a genuinely strong signal. A thin, months-out political contract is a rough directional guide at best, and treating it as a precise probability is where most people go wrong.

The most underrated finding in this space is how much profit concentrates among a small number of disciplined participants who provide liquidity rather than chase hunches. That should reset expectations for anyone entering a market hoping for quick, easy edges. My honest read: use market prices as one well-constructed input, weight them by the conditions above, and never mistake a single number for certainty.

— Max

Sources

FAQ

How often are prediction markets correct?

Prediction markets tend to be well calibrated in liquid, short-horizon contracts and in domains with repeated events, such as elections close to voting day, though accuracy drops for thin or long-dated markets. Long-run studies of election markets found them generally more accurate than polls at both short and long horizons, per research from the Iowa Electronic Markets.

Which prediction market is most accurate?

No single platform is universally the most accurate, since calibration depends more on a contract’s domain, horizon, and liquidity than on the platform itself, according to cross-platform analysis of Kalshi and Polymarket data. A liquid, short-dated contract on one platform can outperform a thin, long-dated contract on another.

Can you actually make money on prediction markets?

Some traders do, but profits concentrate heavily among a small group who typically provide liquidity through limit orders and follow repeatable research processes, while most participants break even or lose money over time. Realistic expectations and disciplined bet sizing matter more than confidence in any single forecast.

What is the most accurate prediction ever?

There is no single documented “most accurate prediction” in the research this article draws on, so any specific claim would be unverifiable. What the evidence does support is that well-calibrated, liquid markets close to resolution consistently produce reliable probability estimates, which is a more useful benchmark than chasing one standout example.

Made with BabyLoveGrowth’s technology