Why Prediction Markets Got the Brazilian Election So Wrong

Prediction markets got the results of Brazil’s presidential election wildly wrong, with Polymarket putting Luiz Inácio Lula da Silva’s chances at just 20% before election day, when he ultimately won out.

Experts say the whiff is the result of Brazil banning prediction markets in the country and a wave of retail investors diluting the expertise of the money on the line.

Luiz Inácio Lula da Silva, commonly referred to as Lula, beat right-wing rival Flávio Bolsonaro in the Brazilian presidential election on Sunday. Bolsonaro came out ahead after the first round of voting at the start of October, but hadn’t reached a majority, and a second round of votes was required. That second round is what Lula won, with a majority this time, and he will serve his fourth term as president.

Going into election day, Lula hovered around a 20% chance of victory with Bolsonaro sitting at 80%, per Polymarket. Even once polls and results started to roll out, one Polymarket whale told Decrypt that the market was slow to push up the odds.

“Sharps [a term for professional, high-stakes predictors] knew the market was mispriced, probably closer to 30-40% on [event] day. Mispricing was most evident during live trading, Lula was nearly a lock by 6:15 pm ET when Lula ‘Yes’ was 60c,” pseudonymous Polymarket whale FhantomBets said. “Retail traders were so against Lula that you could get unlimited size at 60-80c when Lula had [essentially] already won.”

Prediction markets settle at $1, with prices below that—such as $0.60 on a “yes” bet for Lula in Brazil’s election—providing implied odds of that event occurring.

The Polymarket trader added that many whales, including himself, were selling off long-term positions to get liquid capital so they could add more size to the Brazil market. Essentially, FhantomBets is claiming that retail traders with less prediction market experience inflated Bolsonaro’s chances right until the very end.

Luke Nolan, senior researcher at CoinShares, pushed back on this notion and said Polymarket results reacted as polling results were released throughout the night.

“I don’t think more users are inherently a problem. They can bring liquidity and different perspectives, including local knowledge. What matters is whether participation brings different information, and whether traders can act on it,” Nolan told Decrypt. “I would be wary of explaining the result simply as professionals being right and retail being wrong.”

Instead, Nolan suggested Brazil’s ban on prediction markets is likely a bigger reason why the market was so harshly mispriced. Einar Knobel, CEO and co-founder of sportsbook and prediction market firm TxODDS, agreed.

“The price was set mostly by people outside the country,” Knobel told Decrypt. “They were trading the first-round result, poll headlines, and the story built around them.”

Brazil issued a sweeping ban against prediction market platforms in April, citing investor protection concerns and the rise of gambling addiction in the nation. Finance Minister Dario Durigan said the platforms violated betting regulations and lacked proper regulatory oversight.

Knobel added that liquidity was also relatively thin, with the market attracting just 4.5% of the volume that the 2024 U.S. presidential election garnered. “With fewer participants and almost no local knowledge in the book, little was there to correct the price,” Knobel said.

Nolan believes predictors put too much weight on Bolsonaro’s first-round performance, underestimated the increase in voter turnout for the second round, and assumed how voters would transfer based on the eliminated candidates.

“That said, Lula was priced at around 20% before the vote,” Nolan said. “That is still some form of meaningful chance of winning. I think there are good reasons to question whether the market underestimated him. Ultimately, something that has an 80% chance of happening does not happen 20% of the time.”

The mispricing of the Brazilian election was a significant upset, but Nolan cautioned against seeing it as a massive failure. He said prediction markets continue to price things better than traditional betting platforms—pointing to Betfair giving Hillary Clinton an 83% chance the day before the 2016 U.S. election.

“Across enough events, outcomes given a 20% chance should happen roughly 20% of the time. We should also compare their performance with other forecasts available at the same time,” Nolan told Decrypt. “The concern would be a persistent pattern of overconfidence or a failure to incorporate relevant information, rather than an individual upset.”