Prediction markets got Brazil's presidential election wrong, and not by a little. Polymarket and Kalshi, the two largest exchanges where traders buy and sell contracts on real-world events, gave conservative candidate Flávio Bolsonaro roughly 80%, about 4-in-5, to win the October 25 runoff. Luiz Inácio Lula da Silva, the left-leaning incumbent, won anyway.

The miss matters because money moved on it. Brazil's currency, the real, and the Ibovespa, the country's benchmark share index, both rallied in the days before the vote as traders read the market's confidence as close to settled, and both gave the gains back once the result came in. It matters more because the same structural weakness that produced the number is now baked into US midterm contracts that resolve next week.

Build your own theme park, free in your browserAd Advertise here

Polymarket is a crypto-based exchange where users trade $1-settled event contracts in USDC, a dollar-pegged stablecoin, and a contract's price reads directly as the crowd's implied probability. Kalshi is its US-regulated counterpart, and its Brazil market was smaller still. On paper, both are machines for turning many small opinions into one sharp number. Brazil showed what happens when the machine runs on the wrong opinions.

The number climbed as the race tightened

Through the first-round vote on October 4, which Bolsonaro led without clinching a majority, Polymarket priced his chances at about 64%. Over the next three weeks the number only rose, reaching roughly 80% by runoff day. That is the opposite of what the ground was telling anyone following the campaign, where the contest was narrowing, not breaking open.

When Lula won, the contracts settled to zero for everyone holding the Bolsonaro side. The real and the Ibovespa, which had climbed on the expectation of a market-friendly conservative win, reversed. The people hurt were not only traders on the exchange but investors who had read an 80% probability as a near-certainty and positioned real portfolios around it.

A market is only as smart as who trades it

The core problem was depth. Polymarket's Brazil election contracts carried roughly $249 million in total volume across all the candidate markets. The 2024 US presidential race, by comparison, drew about $3.7 billion on the same platform. Brazil's entire multi-candidate event was a rounding error next to it.

Worse, the decisive move happened on an even thinner slice. Between the October 4 first round and the October 25 runoff, the window in which Bolsonaro's implied probability climbed from 64% to 80%, under $32 million changed hands. A price that the whole world read as the verdict of the crowd was set by a very small crowd indeed. A market is only as well informed as the people trading in it, and here there were not many of them.

The locals were locked out

The people trading it were mostly not Brazilian. In April 2026, Brazil's Finance Ministry, the Ministério da Fazenda, moved to block Polymarket, Kalshi and roughly two dozen other platforms on investor-protection grounds, and the telecoms regulator Anatel carried out the takedown. In September the government broadened an existing ban on such online platforms. By election season, a Brazilian who wanted to trade a contract on their own country's election largely could not.

That left the pricing to outsiders. Einar Knobel, chief executive and co-founder of the prediction-market platform TxODDS, argued the ban handed Brazil's election market to people with no vote in it and often little feel for the race, many of them US-based crypto traders. The wisdom of crowds depends on the crowd knowing something local. A crowd watching from another hemisphere, reading the same English-language posts, does not have that edge, and the price it sets inherits the gap.

When a bigger crowd was invited in

Some of that crowd arrived with an audience. On September 20, Elon Musk replied on X to a since-deleted post from Geiger Capital, a markets account, highlighting Polymarket's pro-Bolsonaro number. Musk, who has publicly backed the Bolsonaros and the wider Latin American right, may have drawn still more US traders toward the contract. Whether his post measurably moved the price cannot be proven, but the episode captures the dynamic: a thin market, a famous amplifier, and a number that traveled far faster than any Brazilian polling booth.

The polls told a closer story

Conventional polling never saw the runaway the market described. A BTG Pactual/Nexus poll released on October 24, the day before the vote, put Bolsonaro at 50.5% and Lula at 49.5%, a gap inside the 2-point margin of error. A Datafolha poll two days earlier, on October 22, had the two tied. By every traditional read the race was a coin flip, which is roughly what it turned out to be.

Election day itself was messy in ways a probability model could not capture. Cristiano Zanin, a justice on Brazil's Supreme Federal Tribunal (STF) and once Lula's personal lawyer, issued a unilateral ruling that extended voting by two hours. None of that late noise showed up in a contract still priced at 80%.

Was it a market failure, or just a 20% that happened?

Not everyone reads Brazil as proof the markets are broken. Luke Nolan, head of research at the digital-asset investment firm CoinShares, cautioned against the tidy "professionals beat the public" narrative: an outcome the market prices at 20% is supposed to happen about one time in five, and one-in-five events do occur. On that view the price was defensible and the result was simply the tail landing.

Others say the smart money was never that confident. FhantomBets, a pseudonymous large Polymarket operator, claimed sharper traders had Lula closer to 30-40%, well above the headline-implied 20%, and that the visible top-line number masked a more divided book underneath. Both readings can be partly true: a genuinely close race, priced too hard in one direction by a crowd that was small, foreign and nudged.

The integrity questions are mounting too

The Brazil miss lands while prediction markets are already under a harsher light for how some traders win. US prosecutors have charged a soldier with using classified information about the operation to capture Venezuela's Nicolás Maduro to take winning positions on Polymarket, and separately allege an engineer at Alphabet, Google's parent, used confidential company data to trade on Google's "Year in Search" results. Those are integrity cases, not pricing ones, but they feed the same question regulators are starting to ask out loud: how much should anyone trust a number these markets produce?

The warning for next week

The forward message is blunt. US midterm contracts, which resolve in early November, are in several cases even thinner than Brazil's were, and thin markets distort for exactly the reasons Brazil just demonstrated. A commanding-looking probability on control of Congress may rest on a fraction of the volume its confidence implies.

Democrats Sweep99%AnoNe
D Senate, R House1%AnoNe
Democratic PartyDemocratic Party99%AnoNe
Republican PartyRepublican Party1%AnoNe
Democratic PartyDemocratic Party99%AnoNe
Republican PartyRepublican Party1%AnoNe

The practical takeaway for a trader is not to abandon these contracts but to read the depth behind the price before trusting it. Check how much volume sits behind a 70% or an 80%, ask who is likely trading it, and treat a confident number in a shallow market as a hypothesis, not a result. No outcome next week is a foregone conclusion just because a contract says so.

The Brazil contract did not fail to read Brazil. It read, accurately, the small and mostly foreign group of traders still allowed to touch it after April, and they were never the electorate. That is the uncomfortable lesson heading into next week: a Polymarket price is a precise measurement of its own crowd, and when the crowd and the voters are different people, the number is measuring the wrong country.