Lavareda's predictions for 2026 are correct in 4 out of 5 election forecasts.

Political scientist and sociologist Antonio Lavareda reviewed five electoral predictions he published in Folha de S. Paulo about a year before the 2026 Brazilian presidential election. Four predictions came true; one did not.

The article, published on Globo Online, examines the effectiveness of structural, conjunctural, and hybrid predictive models applied to the election. Lavareda is the honorary president of Abrapel and a researcher at Ipespe, UFPE, and IDP.

Before presenting the analysis, the author warns about the use of political betting platforms such as Polymarket and Kalshi. Political or economic groups can place coordinated bets to artificially alter odds and create the false impression that a candidate has advanced or become the favorite. Since these values ​​are reproduced by the press and circulate on social media as objective probabilities, manipulation can become an instrument of influence over voters, donors, parties, and investors. Lavareda emphasizes that these platforms should not be confused with scientific forecasting models.

The five predictions

The initial prediction, that three candidates would reach double digits in the polls, did not materialize. Lavareda based his prediction on the observation that, in the 13 direct elections held after World War II, at least three candidates had achieved this result in 77% of cases. Augusto Cury reached 11% in a Nexus-BTG poll released this Tuesday (August 31st), but on the same day, the news was dominated by Minister André Mendonça's accusation regarding conversations between Daniel Vorcaro and Minister Alexandre de Moraes. From then on, Cury declined in each poll, and the race became more polarized.

The second prediction, that there was a reduced chance of victory in a single round, was confirmed. In the historical series analyzed, only 23% of elections resulted in two candidates exceeding half of the valid votes: Eurico Dutra, the record holder with 55%, and Fernando Henrique Cardoso in his two candidacies, with 54% and 53%.

The third prediction also came true: the winner of the first round would obtain approximately 47% of the valid votes. Flávio Bolsonaro registered 47,03%. Five of the six presidential elections of the 21st century confirmed the pattern; the percentages for the first round were: Lula in 2002, 46,44%; Lula in 2006, 48,61%; Dilma in 2010, 46,90%; Bolsonaro in 2018, 46,03%; and Lula in 2022, 48,43%.

The fourth prediction, that the election would not be decided by the issue of public safety, came true and was classified by Lavareda himself as the most counterintuitive and risky prediction. Since the beginning of 2025, Quaest surveys had placed security and violence at the top of Brazilians' concerns, and much of the media began to treat the issue as potentially decisive. In the last Ipespe/Abrapel survey, released 48 hours before the election, the issue was mentioned by only 6% of the sample as the most considered in their voting choice, ranking fifth among 13 items; among Flávio Bolsonaro's voters, the rate reached 10%, in third place. Lavareda argued that the Brazilian federal structural model, which before the approval of the PEC on security fragmented police action among 27 state corporations, dilutes the perception of responsibility and reconfigures the weight of the issue in the electoral process.

Hybrid model and end result

The fifth forecast, of a 72% chance of reelection for the incumbent, also came true. The hybrid model used combined political variables, such as the performance of the ruling party, government approval, lead in polls, and rejection of the winner, with economic variables, including Gross Domestic Product (GDP), unemployment, inflation, the Selic rate, real interest rates, and expectations about the economy. The macroeconomic signals were relatively benign. Based on Bayesian logistic regression, the model arrived at the 72% projection.

Scandals involving Master and the Supreme Federal Court (STF), which took center stage in the first round, even managed to shift the odds in favor of the opponent. After the second round, an updated model incorporated data from the latest Quaest poll, which showed Flávio Bolsonaro with only a one-point advantage, government approval surpassing disapproval by two points, and the challenger's rejection rate three points higher than the president's. The structural model indicated a 56% chance for Lula; the conjunctural model indicated a technical tie, with 51%; and the hybrid model placed him as a slight favorite, close to 54%.

By retrospectively examining the ten presidential elections in the available historical series and adopting as a certainty the indication of a probability greater than 50% for the outcome, the structural and conjunctural models were each correct in nine out of ten contests. The hybrid model was correct in all of them.