Are Prediction Markets Better Than Pundits? Here’s what the data says

Are prediction markets better at forecasting outcomes than experts, journalists, polls and conventional forecasting models?

They have become an increasingly popular way of analysing everything from elections and sporting events to economic data and major corporate decisions.

But as their influence grows, the question remains: are prediction markets better at forecasting outcomes than individual experts?

Prediction markets work by turning money into a price. Anyone who thinks a contract is mispriced can buy or sell it, and the price settles wherever the weight of capital says it should. That is a genuinely different mechanism from a poll or a pundit, and it carries a genuinely different weakness: a price reflects the money behind it, not the number of people behind it. Those are not the same thing, and in 2024 they came apart in public. Polymarket confirmed that a single French trader, operating four accounts, had staked around $45 million on a Trump win and was among the largest holders on that side of the book (Semafor, 24 October 2024). A market can be right and concentrated at the same time, and it is worth holding both facts together before asking whether markets beat anybody.

A well-calibrated market is one where its probabilities broadly match what happens in reality over a large number of events.

So, what does the data actually tell us? The most recent research is less flattering than the popular framing, and more interesting for it.

Prediction markets vs pundits

Firstly, it is right to acknowledge this is somewhat of an unfair comparison. Punditry isn't a standardised system. One commentator might confidently predict an outcome based on their experience, while another may disagree based on theirs.

This can be particularly common in sports, where there is often a commercial demand for debate and disagreement. That makes it difficult to measure whether pundits are actually better or worse forecasters than prediction markets.

One study examining 452 expert forecasts across the four US presidential elections from 2004 to 2016 found that the typical expert's vote-share forecast was 7% of the error less accurate than a simple polling average from the same day: a mean absolute error of 1.6 percentage points against the polling average's 1.5 (Judgment and Decision Making). The gap is real but narrow, and the paper is harder on experts elsewhere: it found they follow the polls rather than sufficiently harnessing the other information available to them.

It is only fair to note what the same study found in the experts' favour. A majority of them, 62%, correctly predicted the direction in which the polls would be wrong. That is a harder task than matching a polling average, and it is the clearest sign in the paper that experts were adding something rather than echoing.

That by itself doesn't prove prediction markets are better than pundits. It does, however, highlight the potential value of aggregated forecasting approaches.

A fairer question is whether markets can consistently produce well-calibrated probabilities over a sustained period.

Prediction markets vs polls

This is a much fairer comparison than the one above, because both polling and prediction markets attempt to quantify the likelihood of an outcome.

The 2024 US presidential election provides a useful case study. Prediction markets moved significantly during the campaign, sometimes producing probabilities that differed substantially from polling averages. By 19 October 2024, around three weeks before voting day, Polymarket gave Donald Trump a 60% chance of winning while The Washington Post's polling average showed Kamala Harris leading nationally by two percentage points (The Washington Post).

Trump ultimately won, but one election isn't enough to establish that prediction markets are better than polls. We have written separately about what the 2024 election taught us ahead of the 2026 midterms, and the same caution applies here: a correct call is not a demonstrated method. The more accurate question is whether markets consistently produce better-calibrated probabilities across a large number of events.

What the largest recent study actually found

Here the popular framing and the research part company, and the most striking number concerns the venue this site covers most.

A study of more than 2,500 political prediction markets across the Iowa Electronic Markets, Kalshi, PredictIt and Polymarket, covering the final five weeks of the 2024 campaign and more than two billion dollars of transactions, found that Polymarket was the least accurate of the four. 67% of its markets predicted outcomes better than chance, against 78% on Kalshi and 93% on PredictIt (Prediction Markets? The Accuracy and Efficiency of $2.4 Billion in the 2024 Presidential Election).

The paper's own conclusion is not that markets are good forecasters. It is that these findings "challenge the view that prediction markets necessarily efficiently and accurately aggregate information about political outcomes".

The accuracy figures are only half of it. The study also found little evidence of efficiency, which is the more damaging result for anyone treating a market price as a settled consensus. Prices for identical contracts diverged across exchanges. Daily price changes were weakly correlated or negatively autocorrelated. Arbitrage opportunities peaked in the final two weeks before Election Day, when attention and volume were at their highest.

That last point matters for the question this article is asking. A market that can be arbitraged is, by definition, not aggregating information cleanly. If the same contract is worth different amounts in two places, at least one of those prices is wrong, and the mechanism that is supposed to correct it is not doing so fast enough.

Prediction markets vs forecasting models

Like prediction markets, forecasting models produce measurable probabilities that can be assessed against the eventual outcome.

The 2024 election again provides the clearest example. FiveThirtyEight's final forecast, published on 5 November 2024, gave Kamala Harris a 50-in-100 chance of winning the Electoral College and Donald Trump 49-in-100, which is a coin flip in all but name (538). Prediction markets were considerably more confident in Trump: Bloomberg reported his Polymarket probability at 66.4% on 29 October 2024, a week before the vote.

Trump ultimately won, so the market's read proved closer on this occasion. That does not make prediction markets automatically better than traditional forecasting models. The two approaches process information differently and can produce very different assessments of the same event, and a single election cannot separate a better method from a luckier one.

It is worth being precise about what that Bloomberg report actually said, because it is usually cited in markets' favour. The segment was titled "1% of Polymarket Bettors Are Boosting Trump's Odds". The market's number was closer to the result, and the market got there through a concentrated set of positions rather than a broad consensus. Both of those are true, and an account that gives you only the first is not describing what happened.

Do prediction markets get better as the event gets closer?

A prediction market six months before an event is answering a very different question from one operating minutes before the outcome is known. Both might show a 70% probability, but the information behind that price can be dramatically different.

Unlike a static forecast, a prediction market is continuously updated as traders respond to new information. Market prices are designed to reflect the market's current view of the likelihood of an outcome, and traders can change their positions as new information emerges before an event is completed.

That means a 70% probability six months out reflects the information available to traders at that point; a 70% probability six minutes before an event may incorporate a much larger and more immediate set of information.

A new poll, injury, earnings report, endorsement, weather forecast or piece of breaking news can change the information available to traders, and therefore change the market price.

The 2024 study complicates the intuition that closer always means better, though. Arbitrage opportunities were at their widest in the final fortnight, which is precisely when most people were looking at the prices. Converging toward the outcome and getting there efficiently are different things, and only the first is guaranteed by the clock.

This is also why how a probability moves can be just as interesting as the probability itself. A market moving from 50% to 70% tells us that traders have collectively reassessed the likelihood of an outcome.

Understanding what caused that move can reveal what information the market considers most important.

So, can prediction markets beat the pundits?

What the data shows is that it is hard to say conclusively, and that the honest answer has become less flattering as the research has improved.

Prediction markets can outperform individual experts and conventional forecasting approaches in some situations, while sophisticated forecasting models or expert teams can perform better in others. There is no single forecasting method that consistently gets every outcome right. The largest recent study of political markets found the most-used venue getting two thirds of its markets right, prices that disagreed across exchanges, and arbitrage widening exactly when the audience was largest.

What sets prediction markets apart is not that they are reliably more accurate. It is that they produce a continuously updated, publicly visible probability with money behind it, which is a different kind of information from a confident prediction by a single pundit. Read with the concentration and liquidity caveats attached, that aggregate view is genuinely useful. Read as an oracle, it is not what the evidence supports.