At 10:36 AM ET on Saturday, Polymarket posted this on X:
“BREAKING: Flávio Bolsonaro surges in Brazil’s presidential election odds, with less than 24 hours until voters head to the polls. 82% chance.”
33 hours later, Luiz Inácio Lula da Silva won a fourth term. 51.8% to 48.2%, per Brazil’s electoral court.
Polymarket wasn’t alone. At that moment Kalshi had Flávio at 82.5% and Polymarket US had him at 81.5%. Two CFTC-regulated venues and one offshore exchange, all pricing the same winner. All wrong.
I’m willing to bet you’ll see a lot of takes this week about prediction markets “getting it wrong.” Some will be lazy. An 80% favorite should lose about one time in five, and one upset doesn’t break a forecasting tool.
But this one deserves a closer look, because the miss was big, the money was big, and the reason it happened says a lot about who actually sets these prices.
How Big Was the Miss?
We pulled the full price history for every national presidential election Polymarket has run since 2020 and scored each one the same way: the winner’s price 24 hours before the market passed 90% for good. That’s roughly what the market believed going into the vote.
Lula was at 16.5%.
That’s the lowest pre-result price for an eventual winner in Polymarket’s presidential history:
Brazil 2026, Lula: 16.5% ($243M traded on the event)
Senegal 2024, Faye: 21.5% ($0.1M)
Poland 2025, Nawrocki: 36% ($129M)
Romania 2025, Nicușor Dan: 47.5% ($234M)
US 2024, Trump: 60% ($3.7B)
Senegal is the only race close, and it traded about a thousandth of what Brazil did.
The obvious comparison is 2016. On election morning, Betfair had Donald Trump at roughly 19% to 21% (Business Insider). The PredictWise aggregator had him closer to 12%. Depending on which number you use, Brazil is slightly bigger or slightly smaller. Either way, it’s the same tier.
This is not the biggest election miss ever. Brexit still holds that title. Betfair priced Remain as high as 94% at polls close on referendum night (SBC News). But for a presidential race, in the prediction market era, Brazil is the new benchmark.
The Polls Said Coin Flip While The Market Said Four to One.
What is most interesting here?
The polls didn’t miss this badly.
The night before the vote, Al Jazeera reported the candidates “polling at 50.” Reuters called it “too close to call.” AtlasIntel’s final poll had Flávio slightly ahead. Lula won by 3.6 points. The polls weren’t perfect, but they told you it was close.
The market, meanwhile, had Flávio as a 4-to-1 favorite.
So what was the market seeing that the polls weren’t?
Mostly, history and momentum. Flávio won the first round 47.0% to 45.2%. The Economist noted on Saturday night that no candidate who lost the first round of a Brazilian election had ever come back to win the runoff.
From first-round night until polls closed, Polymarket never had Flávio below 74%.
That’s a reasonable prior. It just isn’t a 4-to-1 prior.
Who Was Actually Setting the Price?
This is where it gets interesting, and where I think the real lesson is.
In April, Brazil’s National Monetary Council passed Resolution 5.298, banning political event contracts. Polymarket is blocked in the country (Teletime). A VPN gets you in, but the millions of Brazilians with the best feel for their own election were, legally speaking, locked out of the market predicting it.
On paper, at least, the price was set from outside Brazil.
And outside Brazil, the noise was loud and one-sided. On September 23, Matt Gaetz posted “Flavio can save Brazil. This is THE big prize in the Americas.” La Derecha Diario, the Argentine outlet, ran at least 45 posts touting Flávio’s Polymarket odds since April, including one identical post from its Argentina, Mexico, Colombia and Ecuador accounts on first-round day (our research, BI-20261014-PER). Pro-Trump accounts shared the 82% number on Saturday night as proof the race was over.
I wrote last year about reflexivity in prediction markets: the price doesn’t just reflect belief, it can shape it (Poly Pump Fake). Brazil is a clean example of the risk. When the people closest to an event can’t trade, and the people who can trade are mostly reading the same feeds, a market stops aggregating information and starts aggregating a mood.
The 20 Minutes the Tide Turned
Sunday night (all times ET):
4:00 PM. Polls close. Lula is at 24%.
4:19 PM. Brazil’s Supreme Court extends voting two hours after transit failures. The count pauses after the first returns. Lula climbs to about 40% by 6:00.
6:06 PM. The extended polls close and the count resumes.
6:08 to 6:09 PM. With 1.6% counted, Lula leads by 4.3 points. All three venues cross 50% inside two minutes.
6:25 PM. Lula peaks at 84% to 86% on all three venues, up from about 45% when the count resumed. Two minutes later, all three are back at 66%.
6:49 PM. With 15% counted, Flávio retakes the lead, 50.1 to 49.9. Lula’s price drops to the low 60s.
6:52 PM. AtlasIntel’s CEO calls the race for Lula.
7:39 to 7:42 PM. All three venues pass 95%. By 8:15 PM Polymarket has him at 99.9%.
Then came the wobble. Brazil’s early count leaned toward Flávio on Sunday, so his lead with 15% in meant less than it looked. Anyone who sold Lula in the low 60s watched it go to 99 about an hour later.
Every Venue Moved Together
One question we get a lot: do the venues actually agree, or is it Polymarket and everyone else?
On Sunday night, the three venues crossed 50% within two minutes of each other, 75% within two minutes, and 95% within three.
They peaked in the same two minutes and fell back together. None of them led. Polymarket US kept pace on a much thinner book, with $1 fills printing 15 cents apart in the same minute.
Did Anyone Know?
Every big upset gets the same question, so we checked.
We pulled every Polymarket fill on the two main runoff markets (151,000 of them through 8:25 PM ET Sunday), Kalshi’s large trades and the Polymarket US tape. In every window before the count resumed, the aggressive money on Polymarket leaned toward Flávio, by $100,000 to $570,000 per window. Kalshi’s large trades leaned the same way. The big Lula money arrived only after results were public.
The largest Lula winners built their positions days and weeks earlier, at market prices around 20 cents. One of them sold out during the count, at 70 to 85 cents. Someone who knew the result wouldn’t do that.
Meanwhile, a group of four accounts with near-identical books, all heavy on Flávio since August, held into the count and lost about $836,000 by our math.
No sign of inside information. Just a market that was confident, and wrong.
What This Means
Prediction markets are still the best real-time read we have on most questions. Once the count started on Sunday, the market went from wrong to right in about 20 minutes.
But going in, the market was only as good as the people allowed to trade it. In Brazil, the people best placed to read the race were barred from trading it, and the loudest voices around the market were cheering one side.
That’s the part I’d want every venue, and every reporter quoting an 82%, to think about before the next one. The US midterms are eight days away.
We replayed all 72 hours, minute by minute, across Polymarket, Kalshi and Polymarket US. [Embed the clip or link the replay.]
Sources
Polymarket, “Flávio Bolsonaro surges...” (X, Oct 24, 2026):
Result, Lula 51.8% / Flávio 48.2% (TSE via Poliwave): https://www.poliwave.com/br/live/presidente-2026
France 24, Lula wins by a narrow margin: https://www.france24.com/en/americas/20261025-live-voting-under-way-across-brazil-in-tight-presidential-run-off-between-lula-and-bolsonaro
CNBC, Lula wins runoff: https://www.cnbc.com/2026/10/26/brazil-lula-win-presidential-election-runoff-vote.html
Al Jazeera, “polling at 50” (X, Oct 24):
The Economist, first-round losers (X, Oct 24):
Oliver Stuenkel on AtlasIntel’s call (X, Oct 25):
Al Jazeera, Supreme Court extends voting (X, Oct 25):
AP, count resumes (X, Oct 25):
Matt Gaetz, “Flavio can save Brazil” (X, Sep 23):
Teletime on CMN Resolution 5.298 (Apr 27, 2026) [Dan: add the article URL from the PER case file]
Business Insider, 2016 election-day odds: https://businessinsider.nl/hillary-clinton-odds-winning-election-trump-2016-11/
WSJ, betting markets and notable exceptions: https://www.tovima.com/wsj/betting-markets-tend-to-get-elections-right-with-some-notable-exceptions
SBC News, Brexit polling-day odds: https://sbcnews.co.uk/?p=32335
City AM, Betfair Remain probability on referendum night: https://www.cityam.com/eu-referendum-implied-probability-remain-slashed-betfair/
BETINT research: Polymarket CLOB price history (34 national presidential races), Kalshi KXBRPRES minute candles, Polymarket US gateway quotes, Polymarket Data API fills; case BI-20261014-PER.