Market, poll or pundit: who was actually right?

On the 2024 US presidential race the market leaned toward the eventual winner and the leading poll model did not. Polymarket had Trump at 58 percent on 4 November 2024, while the 538 model's final forecast gave Harris 50 in 100 and Trump 49 in 100. Pundits produced no scoreable number at all. One election does not settle the general question, and the same market data set looks much worse when you widen the sample.

Which method was closest in 2024?

Method	Produces a number?	Public record per forecast?	Updates in real time?	Money required?	Trump win probability, 4 to 5 November 2024
Poll average model (538)	Yes	Yes	No, published at intervals	No	49 in 100
Prediction market (Polymarket)	Yes	No, prices only	Yes	Yes	58 percent
Pundit commentary	Usually no	No	No	No	Not published as a number
insiderz	Yes	Yes, per person	No, each call is locked	No	Not applicable, launched September 2026

Sources for the two numbers: Fortune, 4 November 2024, recording Polymarket at 58 percent for Trump at 1:35 pm ET on 4 November 2024, and 538's final forecasts, ABC News, 5 November 2024, which put Harris at 50 in 100 and Trump at 49 in 100.

Read the table as a description of what each method offers, not as a ranking of skill. A 58 percent forecast and a 49 percent forecast are barely distinguishable on a single outcome.

What is each method actually doing?

A poll asks a sample of voters who they intend to support, then a model turns many polls into a probability. It measures stated intention at a point in time and it can tell you which groups moved.

A prediction market asks a different question: at what price will you take this side. The number reflects polls plus news plus money plus whatever a trader believes about turnout. It has no opinion about why.

A pundit produces an argument. Arguments carry mechanism, which numbers do not, but they are rarely stated as probabilities and almost never collected into a scored record.

A forecasting platform asks people for probabilities and scores them afterwards. That is the category insiderz sits in.

What did the 2024 studies conclude?

A study of Polymarket against poll averages for the 2024 race examined seven swing states plus national data and concluded that Polymarket was superior to polling in predicting the outcome, particularly in swing states. Polymarket's 95 percent predictive intervals stopped crossing the 50 percent line by mid October 2024, with the clearest divergence from polling between 18 and 29 October 2024, while poll based means never showed Trump as the winner. The authors flagged Michigan and Wisconsin as ambiguous, with both sources near 50 percent (Cutting et al., arXiv:2507.08921, July 2025).

The opposite finding comes from widening the sample beyond the headline race. Across more than 2,500 political markets on the Iowa Electronic Markets, Kalshi, PredictIt and Polymarket in the final five weeks of the campaign, 93 percent of PredictIt markets, 78 percent of Kalshi markets and 67 percent of Polymarket markets priced the eventual outcome better than chance, and prices for identical contracts diverged across exchanges with arbitrage still available in the final two weeks (Clinton and Huang, SocArXiv, 2025).

The reconciliation is simple. The flagship market was good. The average market was not. Anyone quoting "prediction markets called 2024" is quoting one market.

What about the 2026 midterms?

The 2026 US midterm elections are scheduled for 3 November 2026. As of 4 September 2026 nothing has resolved, so there is no scoreboard yet and this section will stay empty until there is one. Any page claiming a 2026 midterm accuracy result today is claiming something that cannot exist.

The comparison worth running afterwards is narrow and specific: for each contested race, the market price on the eve of the vote, the leading poll model's probability on the same date, and the outcome. That is the table this section will carry. The scheduled update is within two weeks of 3 November 2026.

Where do polls still win?

Polls win whenever you need to know why, not just what. They break results down by group, they measure issue salience, and they can detect a change in the electorate that no price will separate from noise. A market can tell you a candidate slipped from 60 to 52. Only a poll can tell you it happened among voters under 30.

Polls also win in places where markets are thin or absent. Most local elections, most referendums outside the largest countries and almost every question about opinion rather than outcome have no liquid market at all.

Where do pundits still win?

Pundits win on mechanism and on questions that have not yet been written down as a market. A good analyst tells you which county reports late, which court is likely to take an appeal, which rule change would flip an outcome. That is information the price consumes, not information the price produces.

The weakness is structural rather than intellectual. Commentary is rarely stated as a probability, so it cannot be scored, so nobody knows which pundits are good. The measured evidence on individual forecasting skill exists but it comes from tournaments, not television: on the ForecastBench human question set, superforecasters reached a mean Brier score of 0.096 against 0.121 for the general public, a gap significant at p below 0.001 (Karger et al., arXiv:2409.19839, revised February 2025).

Why is the market the right bar to beat?

Because it is the only one of the three that is public, numeric and continuous. The poll model publishes at intervals and covers few questions. The pundit publishes no number. The market price exists for every listed event, at every moment, and it already contains the polls.

That makes it the natural benchmark. Beating it is a well defined claim: your probability was closer to the outcome than the price was at the same instant, repeated often enough that luck is not the explanation.

That is how insiderz scores. A call is a yes or a no plus how sure you are, locked the moment it is posted with the time and the Polymarket price for that event frozen alongside it. It cannot be edited or deleted. When the event resolves, the call is scored against the price it was locked at. The leaderboard ranks people on Beats market, Events, Edge and Early. Calls go public after a delay, and followers can ask for live access to see them immediately. There is no money in it.

How this comparison is maintained

The scoreboard uses only numbers with a published date and a named source, taken from the forecaster's own output rather than from later summaries. Where a method does not produce a number, the cell says so rather than guessing one. Elections are added after they resolve, never before. Last checked 4 September 2026, next scheduled update within two weeks of 3 November 2026.

Open events are on Events. The related reading is are prediction markets accurate and when is the market wrong.