Brazil Election Prediction Markets: Why Polymarket Missed
Polymarket gave Lula just a 20% chance of winning Brazil’s election. Learn why the odds lagged, what regulation changed, and how to read market forecasts.

Why Polymarket Misread Brazil's Election - and What It Actually Shows
Prediction markets are often treated like live forecasts: if you put money behind an outcome, the price is supposed to tell you what informed people collectively expect. That shorthand is convenient, but it leaves out some important ground rules. A market can only reflect information that its participants have - and only if they can trade enough to actually move the price.
That's the crux in Brazil's presidential election. Before the second round, Polymarket priced Luiz Inácio Lula da Silva at about a 20% chance of winning, while Flávio Bolsonaro was near 80%. Lula won. At face value, it looked like a big forecasting miss. One Polymarket trader even suggested that retail participants had kept Bolsonaro's odds too high. Others pointed to Brazil's ban on prediction market platforms, thin trading, and limited local insight.
There's also a basic statistical point: a 20% outcome is unlikely, not impossible. One result doesn't prove the system is broken. To understand what happened, it helps to separate the outcome from the mechanism that produced the odds.
This article looks at why Polymarket struggled to price the race, how regulation and liquidity shaped who participated, and what this case can - and can't - tell you about prediction markets as forecasting tools.
Why Polymarket mispriced the Brazil election
Polymarket's prices before election day suggested Lula had about a one-in-five chance of winning. In a contract that pays $1 for a correct "Yes" outcome and $0 otherwise, a price around $0.20 is usually read as an implied probability of roughly 20%. Bolsonaro's price, around $0.80, implied the opposite.
But those prices didn't guarantee what would happen. They were a snapshot of what traders were willing to pay at a particular moment, given the information they had and the conditions for trading. The difference matters: it's the difference between treating the market like an oracle and treating it like a fallible aggregation mechanism.
The first-round result may have carried too much weight
Brazil's election moved to a second round because neither candidate won a majority in the first. Bolsonaro led early, and reporting cited in Decrypt suggested traders may have leaned too heavily on that performance when estimating the runoff.
A second round changes the decision for many voters. People whose preferred candidate didn't advance have choices: support another candidate, abstain, or cast a blank/invalid ballot. The final result depends not just on first-round totals, but on turnout patterns and how those voters transfer their support.
Luke Nolan, a senior researcher at CoinShares, argued that market participants underestimated the increase in turnout and made assumptions about how supporters would shift between rounds. Those assumptions might have sounded plausible from headlines or broad political narratives - but they apparently didn't capture what ended up happening.
A trader's view is not the whole story
A pseudonymous Polymarket whale, FhantomBets, told Decrypt that experienced traders believed Lula's chances were higher than the posted odds. According to the trader, Lula's implied probability was closer to 30-40% by election day, and retail traders kept supporting Bolsonaro's price even as the vote counts started to point toward Lula.
That's a reasonable explanation for one slice of what happened - how a specific person interpreted the order flow. It's not proof that retail traders caused the miss. Nolan also warned against turning it into a simple "professionals were right, retail was wrong" story. More participants can bring liquidity and different perspectives, including local knowledge. The real question is whether participants had relevant information - and whether they could act on it in a way that moved the price.
The price also shifted as results emerged. FhantomBets said Lula's "Yes" contract reached around $0.60 at a point when the trader believed the outcome was already close to decided. That raises a separate issue: markets can react slowly when trading is thin, when participants aren't confident in what new information means, or when the market structure makes it hard to build meaningful positions.
How regulation and liquidity shaped the market
Brazil's regulation is central to understanding who could participate. In April, Brazil issued a broad ban on prediction market platforms, citing investor protection concerns and worries about gambling addiction. Finance Minister Dario Durigan said these platforms violated betting rules and lacked proper regulatory oversight.
Market participants described a practical consequence: many traders were outside Brazil. TxODDS CEO and co-founder Einar Knobel told Decrypt that the price was largely set by people abroad trading around first-round results, polling headlines, and the narratives built from them. That can create an information gap. Public news is accessible everywhere, but local familiarity with day-to-day politics isn't.
Legal access affects the information inside a market
A restriction doesn't automatically make a market "wrong." It changes who shows up. If local residents can't legally or practically trade on a platform, the market may lose people with firsthand knowledge of regional issues, campaign networks, or voter sentiment.
That matters most when information is hard to capture in national polling or in a single headline. Local knowledge isn't automatically better - sometimes it's just more confident about the wrong thing. Still, it can help traders challenge assumptions, spot developments that headlines miss, and take positions informed by what they actually see.
This is a familiar design trade-off in regulated financial and betting products. Stronger oversight can help with consumer protection, but restrictions can also push activity offshore or narrow the range of participants. Where the system ends up depends on enforcement, access to alternatives, and the market's overall size. There isn't a clean rule that more regulation always improves forecasts - or that less regulation always does.
Thin trading makes prices easier to distort and slower to correct
Knobel said the Brazil market attracted only about 4.5% of the volume generated by the 2024 U.S. presidential election market. Volume alone doesn't tell you whether a forecast is "sound," but it does suggest a big scale difference.
Low liquidity can have real effects. With thinner trading, a relatively small buy or sell can move the price more than it would in a deeper market. Traders who think the price is off may also find it hard to take a large position without moving the price against themselves. And if the price can't be challenged easily, it may not respond quickly when new information arrives.
This becomes a feedback loop. Fewer participants can mean less capital available to test a popular view. Weak price discovery can make the market less attractive to informed traders, and limited participation can persist. In Brazil, the ban restricted local participation and the market's volume was comparatively low - conditions that can leave less room for local information to correct the odds.
What the election odds did and did not tell traders
The key takeaway isn't that prediction markets completely failed, or that they're reliable by default. It's that a quoted probability only makes sense in context. A market price is an estimate produced by a specific platform, at a specific time, under specific participation and liquidity conditions.
A 20% probability means the outcome is possible, but less likely than its alternative. If you see 20% assigned consistently across many comparable forecasts, you'd expect it to happen about one time in five over the long run. So it's not surprising that a "20% event" occasionally happens.
The real problem would be something like a persistent pattern - events priced at 20% happening far more (or far less) often than that rate.
A single upset is not enough to measure accuracy
The election result was a significant miss versus Polymarket's final pre-election odds. But one datapoint doesn't tell you whether the market is generally miscalibrated. To judge accuracy, you need a large set of forecasts with recorded probabilities and outcomes, tracked over time.
You also need a fair comparison. Nolan argued prediction markets should be evaluated against other forecasts that were available at the same moment - not against an imagined "perfect" baseline. He pointed to a contrast with traditional betting platforms in the 2016 U.S. election. Betfair gave Hillary Clinton an 83% chance of winning the day before the vote, even though she lost.
That doesn't prove prediction markets are always more accurate than polls, sportsbooks, or expert forecasts. It just underlines why performance has to be measured across many events and against relevant alternatives. A single surprise might be worth investigating, but it's not a final score.
Price is a forecast, not a guarantee or a poll
Market prices can incorporate information quickly when traders have access to it, can act on it, and have enough capital to affect the order book. They can also reflect incentives, risk limits, platform rules, and beliefs about what other participants will do.
So market prices don't line up neatly with a direct measurement of voter preference. A poll samples voters using a defined method. A prediction market reflects what traders are willing to stake on an outcome. They can disagree because they measure different things and operate under different constraints.
For readers, the useful question isn't only, "What probability does the market show?" It's also, "What might be missing from that number?" Local participants, reliable polling, the latest results, enough liquidity, and the ability to trade without major price impact. You won't get the true probability from that - but you can judge how much confidence to place in the market's estimate.
How to evaluate a prediction market forecast
You don't have to trade to decide whether a prediction market price deserves attention. A simple checklist can help you separate a useful signal from a number that looks authoritative while ignoring the conditions behind it.
1. Check who can participate
Start with whether people with relevant knowledge can access the market. For election markets, that often means local residents and people who understand the political context. For financial or regulatory bets, it may mean people close to the policy process or the specific industry being forecast.
Access isn't the same thing as expertise, and local knowledge isn't automatically reliable. The goal is to spot who might be missing. If the platform is restricted in the country being forecast, that should change how confidently you interpret the price.
2. Look at liquidity, not just the displayed probability
A displayed price is only part of the story. Ask how much money is actually available to trade around it, whether a relatively small order can move the price, and whether there's enough depth for informed traders to take positions.
When volume is low, a price can look precise without being robust. In a market with a narrow order book, the implied probability may jump sharply when just a handful of traders act. That doesn't make the price meaningless - it just means your confidence should be more conditional.
3. Identify what information is driving the odds
Sometimes prices move because of genuinely new information. Sometimes they track a widely repeated narrative. In the Brazil case, analysts argued traders leaned too hard on the first-round result and polling headlines, while underestimating turnout and how support would shift in the second round.
When you can, compare market movement to the underlying events. Did the probability change after credible new data? Did it react to a headline without much supporting evidence? Is it pricing a real update - or just building momentum?
4. Use probability language consistently
Don't treat "80% likely" as if it means "certain." An 80% estimate still leaves a one-in-five chance for the other outcome. That's not a technicality you can ignore after a result is known.
For a stronger evaluation, record the market price before the outcome and compare it with polls, expert forecasts, and other platforms available at the time. Over many events, look for calibration: when outcomes are assigned a particular probability, how often do they happen? That's more useful than judging a forecasting system based on its most memorable wins or losses.
This approach also helps you avoid hindsight bias. After an upset, the winner's path can look obvious. Before the result, uncertainty is real. The goal isn't to explain away every miss - it's to distinguish normal probabilistic variation from repeated failures to incorporate relevant information.
Conclusion
Polymarket's Brazil forecast was sharply out of line with the eventual result. But it's not as simple as saying prediction markets - or retail traders - are useless. The market operated under real constraints: local access was limited by regulation, trading volume was low, and some participants may have leaned too heavily on first-round results and polling narratives.
When you read a prediction market, treat the price as a conditional estimate. Check who can participate, how much liquidity exists, what information is moving the odds, and how the forecast compares with alternatives. Then judge performance across many outcomes, not one dramatic election. That's the more reliable way to use market prices: as signals worth inspecting, not answers you accept at face value.