Forecast - 2026 Brazilian presidential election

Two rounds, valid votes

A candidate who wins an absolute majority of valid votes, excluding blank and spoiled ballots, is elected in the first round. Otherwise the top two meet in a runoff. In 2026 the first round is on October 4 and the runoff, if needed, on October 25. President and vice-president run on a single ticket for a four-year term, with one consecutive re-election allowed.

Voting is compulsory from 18 to 70 and optional at 16 and 17, over 70 and for illiterate voters. It uses electronic voting machines: the national result is known the same evening. In 2022 Lula won 48.43% in the first round and 50.90% in the runoff.

Twelve approved candidacies, one substitution pending

The Superior Electoral Court approved twelve candidacies. Pablo Marçal’s (PRTB) was rejected; the party asked to replace him with Leonardo Avalanche, a substitution not yet ruled on at the time of calculation. Polls do not test him separately: if approved, his support is already within the other-candidates line, at most 1%.

What decides the election

Two questions, in this order. Can the incumbent, Lula, pass 50% on October 4? He stands at 44.1% of valid votes. And if not, who wins the October 25 runoff against Flávio Bolsonaro, which the 16 pollsters show as tied?

As in electoral law, figures on this page are shares of valid votes: blank, null and undecided answers are removed from voting intentions.

In the polls, blank, null and undecided answers average 8.6%, from 1.6% to 15.7% by pollster. Four voters in five say their choice is final (Quaest 79%, Datafolha 80%, Nexus 84%); minor candidates' voters are the least settled. In Nexus, 91% of voters say they will vote and 5% probably will. In 2022, abstention reached 20.9% in the first round and blank and null votes 4.4% of voters. A record abstention only changes the outcome if it hits one side more than the other: no published data measures that so far.

A forecasting model is not one more poll. A poll surveys a sample and describes opinion at a given moment. A model surveys no one. It gathers what can be measured, published polls, past election results and the campaign context, and derives a distribution of possible outcomes with their probability. It does not say a candidate will get 44%. It says within which range they will finish, and how likely they are to win.

Why it is sturdier than a single poll

Two pollsters publishing on the same day can differ by several points. The model aggregates every poll registered with the Superior Electoral Court, giving more weight to the most recent, the largest and the pollsters that erred least in the past. It then anchors stated opinion in the vote actually cast at the last presidential election, and takes the context into account: the incumbent's approval, the economy, the issues dominating the campaign.

Past polling errors

Brazilian polls have missed at every presidential election, sometimes by a lot. The model replays those real errors on today's polls, in either direction: it does not assume in advance that one camp will be underestimated. This is what gives each candidate their range and their chance of winning outright.

Two linked rounds

The runoff starts from directly polled head-to-heads, which already contain the vote transfers of first-round electorates. It then moves with the first round: a candidate who surprises on October 4 starts the runoff on October 25 with an edge.

Twenty thousand simulations

A single forecast would give a false sense of certainty. The model therefore replays the election twenty thousand times, varying what is genuinely uncertain: polling error, the gap between pollsters, the movement between the two rounds. The probability shown is the share of simulations in which the event happens.

No probability at 0 or 100%

Uncertainty is this model’s product, never its flaw. The simulations are reproducible, and the model was replayed on the six presidential elections from 2002 to 2022 without any parameter being adjusted afterwards.

Sources

Election results: Superior Electoral Court. Historical polls: Datafolha reports, public tables and Brazilian press for 2002 to 2022. 2026 polls: surveys registered with the Superior Electoral Court and published by pollsters. Context: IBGE, Central Bank of Brazil, pollsters for approval and campaign issues.

Credits

All rights belong to Le Millénaire SAS, 47 rue de Vivienne, 75002 Paris.

Model designed by
Project director: William Thay
Deputy project director: Pierre Clairé
Political analysis and data: Matthieu Hocque
Data analysis, mathematics and validation: Florian Gérard-Mercier