Before Brazil’s presidential vote on Sunday, Polymarket and Kalshi both implied roughly 4-in-5 odds that conservative candidate Flávio Bolsonaro would defeat left-leaning incumbent Luiz Inácio Lula da Silva, which the stock and bond markets interpreted as a potentially positive development for the country’s fiscal sustainability and broader economy. The Brazilian currency and benchmark Ibovespa stock index rallied in the runup to the vote in what may be the biggest misreading of an election outcome in recent memory as Lula overcame the odds and won, foisting massive losses on bullish investors.

The episode won’t do much to burnish the checkered reputations of prediction markets, but the greater concern is the way in which traditional investors credulously took the odds at face value — an error that looms ominously over the US midterm elections next week.

Prediction markets, where users place bets against each other on the outcomes of real-world events, have been under fire for much of the past year. In one well-known case, the Justice Department accused a US soldier of using classified information about the raid to capture then-Venezuelan President Nicolás Maduro to place winning bets on Polymarket. Separately, prosecutors say an Alphabet Inc. software engineer used confidential data to bet on Google’s Year in Search results.

Critics worry that these betting markets have spread access to addictive, casino-style gambling and create an incentive for corruption in sports and politics. A further concern is that markets foster a dangerous feedback loop, in which betting odds drive the tone of content in traditional and social media, influencing public perception and, ultimately, democratic outcomes.

The counterpoint is that prediction markets host many contracts that provide useful information into how crowds of people are thinking about an event, which is valuable for risk-management and useful to members of the public trying to understand the future. They help us process a world of data, in mimicking (however imperfectly) the sorts of predictive model outputs that used to be exclusively available to hedge funds, insurers and others in the business of risk analysis. But the signals from prediction markets can deceive more than they illuminate when contracts are small and illiquid. Also, the markets are only as informed as those participating in them.

In Brazil, there were red flags on both of those issues for weeks. Bear in mind that the Polymarket contracts in question had total volume of around $249 million, compared with the $3.7 billion for the 2024 US presidential election. The Kalshi market was smaller still. And that vastly overstates the actual market activity during the key weeks for financial markets. The perceived odds shifted dramatically from Brazil’s first-round balloting on Oct. 4 through the runoff on Oct. 25, with Bolsonaro’s probability jumping to around 80% from about 64%.

Yet there was less than $32 million in Polymarket volume during that window; much of the trading happened prior to the first round and immediately after the results, and then during the frenzied trading toward the end of runoff day, after Supreme Court Justice Cristiano Zanin issued a unilateral decision to extend voting by two hours (Zanin was Lula’s former personal lawyer, as my Bloomberg Opinion colleague Juan Pablo Spinetto points out.) Locals weren’t involved in the repricing at all.

The Finance Ministry, under the incumbent Lula, had blocked access to Polymarket and Kalshi in April, and the government issued a broader ban on online gambling in September. Outsiders checking Kalshi and Polymarket odds for signals were getting the “wisdom of crowds” from non-Brazilians, including US-based crypto bros. Recall that Elon Musk may have brought more of the latter into the contract on Sept. 20 when he responded to a (since deleted) Geiger Capital post on X about how the Polymarket odds favored Bolsonaro. Musk has a long history of boosterism for the Bolsonaros, and the Latin American right in general.

With all of this, I’m hardly shocked that the Brazil contracts sent inaccurate signals. What’s surprising is how many people failed to see the warnings and used them to price other types of financial assets. A BTG Pactual/Nexus poll published the day before Brazil’s runoff showed Bolsonaro, the son of former President Jair Bolsonaro, with 50.5% support to Lula’s 49.5%, within the 2 percentage point margin of error.

A Datafolha survey published Oct. 22 had Bolsonaro and Lula in a tie. Granted, Brazil watchers had learned to expect the Bolsonaro family to outperform polls, but it’s hard to imagine how anyone could have made the logical leap from those results to 80% odds for Bolsonaro.

There are major differences between Brazil and US markets on Polymarket and Kalshi — not the least of which is the fact that actual voters are participating. Yet investors and voters must heed the lessons of the South American episode.

First, midterm betting markets generally have far less activity than the presidential race. In many individual elections, the depth of these markets is much worse than what was witnessed with Brazil. Second, Americans must be cautious about the assumption that any outcome is a foregone conclusion. That belief may have kept the so-called Bolsonaristas at home Sunday and ultimately helped tip the election to Lula. (Though Brazilians couldn’t participate in betting markets, they still heard about them through local media.) Finally, they should take into account the underlying quality of the market in question, if they rely on betting market odds at all.

The risk of misreading prediction markets is greatly amplified when traditional markets are involved. Though the Brazil Polymarket contracts only saw around $32 million in volume between the first round and the runoff, they helped fuel a political narrative that briefly created around $130 billion in stock market wealth, which subsequently vanished when markets reopened on Monday morning. If it happens again around the US elections, the havoc would reach a whole new level.