Prediction Markets vs Models at the World Cup
Prediction markets favor France while the Opta model favors Spain. We test which forecasts the 2026 World Cup better and how accurate each really is.
By Capital & Compute
Two of the most credible forecasts of the 2026 World Cup disagree about who wins it. As of June 25, 2026, the Polymarket winner market puts France first at about 19%. The Opta supercomputer puts Spain first at 16.1%, with France only third. Same tournament, same teams, two methods, two different favorites.
That gap is the interesting part, and it is not really a soccer question. It is a forecasting question that shows up every election, every product launch, and every model-release rumor: when a betting market and a statistical model point in different directions, which one should you believe? This piece treats the World Cup as the test case, because the data is public on both sides, and pulls out what actually separates the two methods.
The disagreement, in one chart
Line the two forecasts up team by team and the pattern is clear. The market is more bullish on France and Argentina; the model is more bullish on Spain. Everywhere else the two are close.
| Item | Polymarket | Opta model |
|---|---|---|
| France | 19.3% | 13.0% |
| Argentina | 14.9% | 10.4% |
| Spain | 13.9% | 16.1% |
| England | 10.6% | 11.2% |
| Portugal | 7.9% | 7.0% |
| Netherlands | 5.7% | 3.6% |
| Brazil | 5.5% | 6.6% |
| Germany | 5.3% | 5.1% |
| Team | Market (Polymarket) | Model (Opta) |
|---|---|---|
| France | 19.3% | 13.0% |
| Argentina | 14.9% | 10.4% |
| Spain | 13.9% | 16.1% |
| England | 10.6% | 11.2% |
| Portugal | 7.9% | 7.0% |
| Netherlands | 5.7% | 3.6% |
| Brazil | 5.5% | 6.6% |
| Germany | 5.3% | 5.1% |
Note one thing before reading too much into any single number: these two snapshots are not from the same day. The Opta run is its pre-tournament forecast from June 1, 2026, built before a ball was kicked. The Polymarket prices are live, taken on June 25 with the group stage already underway. Some of the market’s extra confidence in France could simply be newer information the static model has not absorbed yet. That timing mismatch is exactly the kind of thing the accuracy debate turns on.
Update (June 30, 2026): That mismatch stopped being hypothetical. On June 29, Paraguay knocked Germany out in the round of 32, drawing 1-1 after extra time and winning 4-3 on penalties, as reported by FOX Sports. Watch what each forecast did with it. The Polymarket market repriced Germany’s title chance to essentially zero as soon as the result was in, because a market reprices the instant news breaks. The Opta model, never re-run since June 1, still carries Germany at 5.1% in the snapshot above, because a static model cannot react until its maker reruns it. That is the market-reacts-fast, model-lags split playing out in a single result, and it is why the live World Cup forecast tracker dropped Germany the moment the result was in and moved the next contender into its place.
How a market forecasts
A prediction market is a forecast assembled out of money. On Polymarket, a contract on “France wins the 2026 World Cup” trades between 0 and 1 dollar, and it pays one dollar if France wins and nothing if they do not. If that contract trades at 19 cents, the market is collectively saying France has roughly a 19% chance. The price is the forecast.
What makes the price informative is that being wrong costs the trader money. Someone who genuinely believes France is underpriced can buy until the price reflects their view, and someone who thinks France is overhyped can sell. The economist’s case for these markets, laid out in Justin Wolfers and Eric Zitzewitz’s 2004 paper Prediction Markets in the Journal of Economic Perspectives, is that this process aggregates scattered private information into a single number, and that the resulting forecasts “are typically fairly accurate” and beat most moderately sophisticated benchmarks. The depth here is real: the Polymarket winner market alone shows roughly $3.1 billion in volume and $478 million in liquidity, so the price reflects a large pool of capital, not a handful of fans.
The weaknesses are just as structural. Markets price longshots poorly, a long-documented favorite-longshot bias where the field of small-probability teams trades a little too high. Thin markets can be moved by a single large bet. A 2025 arXiv preprint by Itzhak Rasooly and Roberto Rozzi, How Manipulable Are Prediction Markets?, ran a field experiment across 817 markets and found that deliberate trades could move prices in ways still visible 60 days later, though markets with more traders, higher volume, and an external probability estimate resisted it best. The deeper the market, the less these problems bite, which is why a multi-billion-dollar World Cup market is more trustworthy than a thin market on an obscure question.
How a model forecasts
The Opta supercomputer takes the opposite approach. Instead of asking people to bet, it builds a rating for each team and then simulates the entire tournament many times over. For its 2026 World Cup forecast, published June 1, 2026, Opta ran the tournament 25,000 times and counted how often each team came out on top. Spain won 16.1% of those simulations, France 13.0%, England 11.2%, and defending champion Argentina 10.4%. The percentage is just the share of simulated worlds a team won.
A model’s strength is consistency. It applies the same logic to every team, has no emotional attachment, and will not overrate the host nation because the crowd is loud. Its weakness is that it is only as current as its inputs. A pre-tournament model does not know about an injury in the second group game, a tactical change, or a key player rediscovering form. It updates when its maker reruns it, not continuously. That is the mirror image of the market’s main advantage: the market reprices the instant news breaks, while the model waits for its next run.
So which one is right?
The honest answer is that you cannot tell from one tournament. A forecast that gives the eventual winner a 16% chance was not “wrong” if that team loses, and a forecast that gave them 19% was not “right” if they win. With a single event, you are looking at one draw from a probability distribution. Picking the winner once is luck as much as skill.
The real test is calibration, measured over many forecasts. A well-calibrated forecaster who says “20%” should be right about one time in five across all the things they call 20%. The standard scoreboard for this is the Brier score, which penalizes confident wrong calls and rewards confident right ones, where lower is better. Run it over hundreds of resolved markets and you can see whether a forecaster’s stated probabilities hold up.
Here is where the popular story gets ahead of the evidence. Platforms advertise impressive numbers: Polymarket has been reported, in coverage such as DeFiRate’s prediction-market tracker, to be accurate more than 90% of the time a month before a market resolves, with a Brier score around 0.084. Treat that as marketing, not proof. Those figures are largely self-reported by the platform, not independently audited, and “accurate 90% of the time” quietly counts every market where the heavy favorite duly won, which is easy. More to the point, I could not find a single peer-reviewed study that puts prediction-market Brier scores and model Brier scores side by side for past World Cups. The clean head-to-head everyone assumes exists does not, at least not in the open literature. Anyone who tells you markets definitively beat models at soccer is reaching past the data.
This is the same trap that shows up in AI, where a single headline benchmark gets quoted as proof one system is better than another. We made that case at length in how AI benchmark scores get gamed: the number is only as good as the way it was measured, and a self-reported number measured by the party it flatters deserves the most scrutiny.
What the disagreement is actually telling you
Step back and the France-versus-Spain split is not a contradiction to resolve but information to read. The model, working from longer-run team strength, sees Spain as the most complete side. The market, pricing in everything up to this morning including early-tournament form and money flowing toward France, has nudged France ahead. When a fast, news-sensitive forecast and a slow, fundamentals-driven one diverge, the gap usually marks recent information the model has not yet ingested. That does not make the market correct. It makes the disagreement a flag for where the live story and the long-run baseline have come apart.
The practical takeaway travels well beyond soccer. For any forecast you rely on, ask what it is made of. A market gives you fast aggregation and a price that moves with the news, at the cost of thin-market noise and longshot bias. A model gives you consistency and a transparent method, at the cost of lag. Neither is a crystal ball, and the most useful read is often both at once: where they agree, confidence is higher; where they split, something recent is in play. We track that live split, market versus model, on the World Cup forecast tracker, and we apply the same do-not-trust-one-number discipline to AI release odds in why the GPT-5.6 leak is real but the date is not.
Frequently asked questions
- Are prediction markets more accurate than statistical models?
- Not provably, at least for the World Cup. Prediction markets are usually well calibrated and react faster to news, and economic research finds they beat many simple benchmarks. But the accuracy figures platforms advertise are mostly self-reported, and no peer-reviewed study has compared market and model accuracy head to head across past World Cups. Where they disagree, the gap is best read as a signal, not proof that one is right.
- How accurate is Polymarket?
- Polymarket has been reported as accurate more than 90% of the time a month before markets resolve, with a Brier score near 0.084. Those numbers are largely self-reported by the platform rather than independently audited, and the headline figure is flattered by all the obvious-favorite markets that resolve as expected. The deeper a market, the more its price reflects real aggregated information.
- What is a Brier score?
- A Brier score measures how good probabilistic forecasts are over many predictions. It rewards confident correct calls and penalizes confident wrong ones, and a lower score is better. It is the standard way to test whether a forecaster who says 20% is actually right about one time in five, which is the real measure of accuracy, not whether they picked a single winner.
- Who do the market and the model favor to win the 2026 World Cup?
- As of June 25, 2026, the Polymarket winner market favors France at about 19%, while the Opta supercomputer favored Spain at 16.1% in its June 1 pre-tournament run, with France third. The two forecasts agree closely on most other teams and differ most on France and Argentina, where the market is more bullish.
Sources
- Polymarket (2026). World Cup Winner market (implied probabilities, volume, and liquidity), accessed via the public Gamma API. Prices as of June 25, 2026. https://polymarket.com/event/world-cup-winner
- The Analyst / Opta (2026). Who Will Win the 2026 FIFA World Cup? The Opta Supercomputer Predictions. Pre-tournament run of 25,000 simulations, June 1, 2026. https://theanalyst.com/articles/who-will-win-2026-fifa-world-cup-predictions-opta-supercomputer
- Wolfers, J., and Zitzewitz, E. (2004). Prediction Markets. Journal of Economic Perspectives 18(2): 107-126. https://www.aeaweb.org/articles?id=10.1257/0895330041371321
- Rasooly, I., and Rozzi, R. (2025). How Manipulable Are Prediction Markets? arXiv preprint. https://arxiv.org/abs/2503.03312
- DeFiRate (2026). 2026 World Cup Odds: Live Kalshi and Polymarket Predictions (secondary coverage relaying Polymarket’s self-reported accuracy figures). https://defirate.com/prediction-markets/world-cup-odds/
- FOX Sports (2026). Paraguay Knocks Germany Out Of 2026 World Cup After Penalty Shootout (match-result coverage of the June 29 round-of-32 game). https://www.foxsports.com/stories/soccer/paraguay-knocks-germany-out-2026-world-cup-after-penalty-shootout