Prediction markets have become the consensus reference for event probabilities. Platforms like Polymarket now carry millions in volume on individual World Cup matches, and their prices are widely treated as the best available estimate of what will happen.
On draws, the model disagrees.
The gap in today's opener
The model gives Mexico vs South Africa, the opening match of the 2026 World Cup, a 27.2% draw probability. The consensus across prediction markets sits closer to 20%. That 7-point gap is the single largest divergence between the model and the market on today's card.
Both sources agree Mexico is the favourite. But prediction markets push Mexico's win probability to nearly 70%, leaving less room for the draw and for South Africa. The model sees a more competitive match: Mexico at 59.5%, the draw at 27.2%, South Africa at 13.4%.
The question is whether this pattern is a coincidence or something structural.
Three reasons prediction markets compress draws
1. Traders think in binaries.
The most natural question about a football match is "who wins?" not "does anyone win?" Prediction markets split a three-outcome event into separate yes/no contracts. The "Will Mexico win?" contract attracts the most attention, the most liquidity, and the most confident opinions. The draw contract gets less of all three.
Lower liquidity means less price discovery. The draw price tends to reflect the residual after the two win outcomes are priced, rather than an independent estimate of how likely a draw actually is. This is a known structural weakness: prediction markets are excellent at pricing binary outcomes and less reliable when they decompose multi-way events into separate contracts.
2. Draws have no narrative.
"Mexico wins the opener at the Azteca in front of 87,000" is a story people want to trade on. "South Africa pulls off the upset" is a smaller but still tradeable story. "0-0 after a cagey first half and a couple of half-chances" is not a story anyone trades on for excitement.
Behavioural finance calls this the favourite-longshot bias: people overweight outcomes with strong narratives and underweight outcomes that feel like nothing happening. In football, the draw is the "nothing happening" outcome, even when it's the second most probable result.
3. Home advantage gets inflated in openers.
Host nations in World Cup openers carry a narrative premium that goes beyond what the data supports. Mexico at the Azteca is a powerful image. But the model, which already factors in a measured home-field advantage, still only gets to 59.5% for a Mexico win. The remaining 9 percentage points that the market adds have to come from somewhere, and they come from the draw and from South Africa.
Tournament openers historically trend cautious. Teams play not to lose their first match. The tempo is slower, the tactical approach more conservative, the first goal comes later. These are the exact conditions that produce draws, and prediction markets consistently underweight them.
South Korea vs Czech Republic tells a different story
The second Group A match, South Korea vs Czech Republic, is a useful control case. The model gives it 40.1% South Korea, 26.5% draw, 33.4% Czech Republic. Prediction market consensus is broadly similar, with slightly less on South Korea and slightly more on the draw.
Why does the draw compression not apply here? Because there is no clear favourite. When a match is perceived as roughly even, traders don't have a dominant narrative to rally around. Neither "South Korea wins" nor "Czech Republic wins" is an obviously exciting trade. The result is that the draw gets priced more fairly.
This supports the structural argument. Draw compression is worst when one team is a heavy favourite, because that's when the favourite narrative is strongest and absorbs the most trading volume.
What the model sees that the market doesn't
The model's draw probability comes from the Dixon-Coles framework, which explicitly estimates a draw inflation parameter from historical match data. This parameter captures a real feature of football: draws happen more often than a simple Poisson model of independent goal-scoring would predict. Teams adjust tactically, matches settle into patterns, and 0-0 or 1-1 are equilibrium outcomes in ways that a pure probability-of-goals model underestimates.
Prediction markets have no equivalent mechanism. They aggregate individual trader opinions, and those opinions are shaped by the biases above. The result is a systematic gap: draws are underpriced, and the gap is widest in matches with a clear favourite.
For today's opener, the model's top three scorelines are 1-0 Mexico (16.9%), 0-0 (14.5%), and 1-1 (13.1%). Three of the five most likely outcomes are draws or low-scoring results. The model sees a match where South Africa's defensive structure under Hugo Broos makes this harder for Mexico than the market expects.
All probabilities in this post are model outputs as of the June 11 frozen snapshot. They reflect what falls out of the Elo ratings, expected-goals data, and group composition when the ensemble runs its Monte Carlo simulation. Full methodology: /docs/methodology/.
The model publishes probabilities, not recommendations. It can be wrong. Terms of Use.