São Paulo Prediction market platforms, such as Kalshi and Polymarket, despite being banned in Brazil, still offer projections about the Brazilian elections. This data circulates in the press and among economic analysts, although the companies themselves recognize that the blockade in the country harms their results.
These sites offer bets on elections, events, games, reality shows and celebrities. The government banned them in April, even before shutting down the regular betting market in September, as it understood that they offered sports predictions without respecting the rules imposed on bookmakers.
Kalshi and Polymarket, however, say that their main product is the statistical projection of what will happen in the future, and the worsening of Brazilian projections is one of the arguments that companies use to try to return to operating in the country, according to spokespeople interviewed by the report.
Unlike a sample survey, such as a traditional election survey, which asks the same questions to a group with a distribution of people similar to that of the entire population, prediction markets group together the guesses that people bet money on.
The idea is that the willingness to pay validates the positioning, said Brazilian Luana Lopes Lara, co-founder of Kalshi, during an event in Sunnyvale, California, accompanied by the report in April.
Although they do not rely on sampling, prediction markets assume high liquidity — that is, enough participants to buy and sell contracts on each candidate's victory. The ban in Brazil generates a market failure because it restricts the participation of interested people.
The volume bet on the Brazilian elections on Kalshi and Polymarket represents just 4.5% of the volume bet on the dispute between Donald Trump and Kamala Harris in the United States presidential elections in 2024, when these platforms were less popular around the world.
In the case of Polymarket, where each movement is recorded on a blockchain, as in operations with bitcoins, the average daily transactions on the future of the Brazilian elections fell 89.6% after the ban on prediction markets in Brazil (considering the 73,568 operations this year). Kalshi does not disclose data on how many people are in each section.
The biggest active electoral bet at Polymarket is in favor of the re-election of President Luiz Inácio Lula da Silva. The user "unknwnfnd" placed US$1.01 million (R$5.26 million) on the PT's victory, when his chances of winning were estimated at 48%.
As the incumbent's chances of victory for Polymarket fell to 41% this Saturday (3), this person would lose US$67,000 (R$349,100) if they sold their contracts. Flávio Bolsonaro (PL) appears ahead, with a 59% chance of success.
This Saturday, Missão's presidential candidate, Renan Santos, who appears in the last round of Datafolha with 3% voting intention, has 17 bets of more than US$ 100 thousand, a number higher than that of Lula (with 8) and that of Flávio Bolsonaro (with 7).
The number of big bets, however, does not correspond to the probability attributed to each candidate. In prediction markets, the chance is determined by the price of the contracts, which varies depending on supply and demand.
The method of prediction markets in the electoral context is the target of criticism from political scientists and statisticians, due to the lack of independent scientific material to validate these projections.
Kalshi and Polymarket, on the other hand, claim that they were the platforms that predicted Trump's victory in the 2024 elections.
Despite liquidity problems due to restrictions in 53 countries, including Brazil, Luana, from Kalshi, said in an interview with Folha that predictive markets are an alternative to electoral polls. “The data is qualitatively superior,” she says.
"Electoral markets are the most important within the predictive segment, as it is notoriously difficult to obtain reliable data on voting trends. Conventional opinion polls face sampling crises, with low response rates, and often lack methodological transparency", he states.
The professor of quantitative methods at the Department of Social Sciences at USP, Glauco Peres, assesses, on the other hand, that the platforms have not proven in independent research their predictive power, the ability to accurately predict future results based on available data and information.
If, on the one hand, research published by the NBER (National Bureau of Economic Research) showed that the tool has an accuracy similar to that of derivatives used in the financial market and opinion polls in forecasting interest rates, on the other, Robert Bartlett, from Stanford University, and Maureen O’Hara, from Cornell University, show the effects of privileged information on topics not directly related to macroeconomics.
In an analysis of events linked to specific companies or people (such as mentions in Netflix conference calls or the presence of celebrities at the Super Bowl), researchers noticed that market makers tend to profit more than the rest of investors, an indication of insider trading, and that there is a bias in buying positive contracts — that is, in bets that something will happen, to the detriment of negative ones.
In the case of electoral research, the sample survey is a reference in several fields of knowledge.
