Prediction
Markets UK

An independent guide to
an uncertain subject.

Our editorial approach
02 / FORECASTS

Are prediction markets accurate? 74%

Evidence from Iowa election markets, economic event forecasts and Brier scores—and why timing and market design matter.

Updated · Research note
The short reading

Some markets have performed well in specific comparisons. Accuracy is not a permanent property of a brand: it depends on the question, time to resolution, liquidity and benchmark.

This is research and context, not a market recommendation or a statement of UK product availability.
At a glance74%of 964 poll comparisons favoured the Iowa Electronic Markets forecast in one historical studyView original source
Measure / 01964poll comparisonsFive US presidential elections through 2004.View original source
Measure / 021.33ptelection-eve absolute errorAverage IEM error reported for the illustrated elections.View original source
Measure / 030.09 → 0.02Brier score as close approachedAround three months out versus near close in a Kalshi-authored analysis; lower is better.View original source
01 / Analysis

What the Iowa comparison really measured

Berg, Nelson and Rietz compared Iowa Electronic Markets forecasts with 964 polls over five US presidential elections. The market was closer to the eventual result in 74% of those comparisons. The study covers a particular election era and a poll baseline; it does not mean every current event market is right 74% of the time.

02 / Analysis

Forecasting rates and inflation

Federal Reserve researchers have examined the information in event-contract prices about economic releases and policy decisions. Their findings show that market-based forecasts can be useful against selected benchmarks. A useful forecast can still miss, and a contract price has to be interpreted with its exact settlement rule and observed trading depth.

03 / Analysis

Why timing changes the score

A Brier score measures how far a probability forecast is from the binary outcome; lower is better. Semafor reported a Kalshi-authored analysis with scores of roughly 0.09 three months before resolution and 0.02 near close. Late certainty is easier to achieve than early insight, and platform-authored studies need an independent reading of methods.

04 / Analysis

Where markets can go wrong

Thin order books, ambiguous questions, selective participation and sudden new information can distort prices. Comparing a market near settlement with a months-earlier poll is not a like-for-like test. Treat prices as one evidence stream alongside reporting, expert analysis and the written contract—not as a guarantee.

Figure / brier-score

Kalshi-reported Brier scores by lead time

Kalshi-reported Brier scores by lead timeHorizontal bar chart. Brier score; exact figures appear in the data table below. MEASURE BRIER SCORE Three months out 0.09Near close 0.02
Chart data: Kalshi-reported Brier scores by lead time
MeasureBrier score
Three months out0.09
Near close0.02
Lower scores indicate more accurate probability forecasts in this study.

Reading note. Platform-authored study reported by Semafor; the chart is not an independent validation.

SourceKalshi banks on its predictions’ accuracy
Method / How to read this

Methodology.

The Iowa paper covers elections through 2004; the Federal Reserve analysis concerns selected economic contracts; the Brier comparison comes from a Kalshi-authored study reported by Semafor. Their measures are not interchangeable, and none describes UK platforms generally.

Reference / Original record

Sources.

  1. 01
    Prediction Market Accuracy in the Long Run (opens in a new tab)
    Research papers

    Funder: University of Iowa academic working draft; see acknowledgements.

    Berg, Nelson and Rietz compare Iowa Electronic Markets forecasts with 964 polls across five US elections through 2004.

  2. 02
    Federal Reserve FEDS paper on prediction-market economic forecasts (opens in a new tab)
    Research papers

    Funder: Federal Reserve Board research series; findings are the authors’ own.

    Study of event-contract forecasts for economic releases and policy decisions; not a blanket accuracy ranking.

  3. 03
    Kalshi banks on its predictions’ accuracy (opens in a new tab)
    Reporting

    Funder: Semafor reporting; the analysed accuracy study was conducted by Kalshi researchers.

    Reports Brier scores of approximately 0.09 three months before close and 0.02 near close. Platform-authored evidence warrants caution.

Follow the original documents for definitions, scope and subsequent revisions. Browse the full source library ↗

Questions / In context

Questions worth asking

01Does a 70% contract price mean the event will happen?

No. It is a market-implied estimate before costs and market frictions. Even a well-calibrated 70% forecast would fail about three times in ten over many comparable cases.

02What is a Brier score?

It is the squared difference between a probability forecast and the binary outcome, averaged across questions. Lower is better, but compare studies with similar question sets and lead times.

03Can prediction markets beat polls?

They have in particular historical comparisons, including the Iowa study. That does not establish a universal advantage across all topics, periods or forecast horizons.