Polymarket gave Luiz Inácio Lula da Silva only about a 20% chance of winning Brazil's presidential election. The left-wing candidate won on Sunday anyway, a result that stunned the prediction market world and raised questions about what truly determines the odds on such platforms.

Prediction markets like Polymarket let traders buy and sell contracts tied to real-world events, with prices updating continuously as new information arrives. Their odds are increasingly cited alongside polls as a real-time gauge of election sentiment, which is why a miss of this size has drawn such scrutiny.

Lula defeated his right-wing opponent, Flávio Bolsonaro, in a second round of voting. A runoff was needed because Bolsonaro led October's first round but failed to secure a majority. In the second round, Lula managed to capture a majority of the votes.

Before the vote, Polymarket put Lula's chances of winning at 20% while pricing Bolsonaro at around 80%. Contracts on prediction platforms settle at $1, so a 60-cent "yes" price implied a 60% probability of a Lula victory. The pre-vote odds therefore treated Lula's win as an unlikely scenario. News of his victory, however, sent the Brazilian stock market falling.

A Polymarket whale blames retail

One of the platform's big traders says the problem was the crowd. FhantomBets, the anonymous pseudonym of a Polymarket whale — trading slang for a participant with outsized positions — stated in a report that veteran bettors realized the odds were wrong and that Lula's actual probability of winning was closer to 30% to 40%.

The disparity became most apparent once votes were counted. "The retail traders were so much against Lula that you could get unlimited size at 60-80c with Lula having [essentially] already won," FhantomBets said, noting that Lula's win was certain by 6:15 pm ET even though his "yes" contract was still priced at just 60 cents. The whale added that he and other traders exited their long positions to enter the Brazilian market.

The ban that thinned the book

The professionals-versus-amateurs explanation does not hold for everyone. Luke Nolan, head of research at CoinShares, said he would exercise caution when interpreting the result as smart traders beating retail. In his view, greater participation would have meant greater liquidity and richer insight into the state of the race.

In trading parlance, a thin book means fewer resting orders on either side of the market, leaving quoted prices more exposed to swings on relatively small flow. Proponents have long argued that prediction markets work best when they pool information from the widest possible set of participants, a premise that sits at the center of the debate over what went wrong in Brazil.

Nolan and Einar Knobel, CEO and co-founder of prediction-market platform TxODDS, both pointed to an obvious reason for the thin book: Brazil's government banned prediction markets in April, citing investor protection and growing problems with gambling addiction. Finance Minister Dario Durigan claimed the platforms had violated betting regulations and had not received any authorization from regulators.

Because local players were locked out, Knobel said the price was formed "by people outside the country" who traded on poll headlines and the first-round narrative. Total volume corresponded to only 4.5% of the activity seen in the 2024 U.S. presidential market.

Was it really a failure?

Nolan disagreed with describing the outcome as a blowout. A 20% probability, he said, is still something that can happen; an outcome carrying those odds should come true once in five tries, and this could simply have been one such occasion.

He also argued the market performed better than traditional betting companies, pointing to Betfair's 83% odds for Hillary Clinton on the eve of the 2016 U.S. election. In Nolan's opinion, though, this was not merely one wrong prediction but part of a pattern of overconfidence that should be alarming.