Brazil's 2026 presidential election: the probabilities
Data as of 2 Oct 2026
Lula leads the first-round polls. Trouble is, for 6 elections running his party's rivals have beaten their polls.
1 The bottom line
- The favourite is Flávio Bolsonaro. Chance of becoming president: Flávio, 71% to 80%; Lula, 20% to 29%. The top end for Lula is our main estimate; the bottom end is the same estimate without one softening assumption about the runoff (see “What the model does not know”). We give a range rather than a single number because the honest answer depends on that assumption.
- The favourite is not the candidate leading the polls. In the first-round poll average Lula is on 44.3% of valid votes and Flávio on 42.1%. But in all 6 past elections, 2002–2022, the Workers' Party (PT, Lula's party) did worse against its rivals than the polls a few days before the vote suggested (table and sources under “How we calculated”). The model carries this lesson over to 2026, so Flávio's chance of finishing first in the first round is 64% and Lula's is 36%.
- The chance of a runoff is 74%. In head-to-head polls of Lula against Flávio, Lula is on 49.4% of the two-candidate vote. Each one-point shift in Lula's share in these polls moves his chance of winning by about 4.8 points: one point more for Lula takes it to 34%, one point less to 24%.
The key assumption in one line. In 4 of 5 past cases, runoff polls taken before the first round overstated PT; we carry this correction over in softened form (on average 2.3 points against Lula). If this time the head-to-head polls do not systematically overstate Lula, his chance is 41%.
2 How to read these percentages
A chance of 71% is not a sure thing. Think of a weather forecast giving the same chance of rain: an umbrella is worth taking, but a dry day is entirely possible. The percentages come from simulation: we “play out” the election 200,000 times, each time adding to the polls an error of the size and direction seen in Brazil before, and count how often each candidate wins.
Our calculation is an estimate based on public data as of 2 Oct 2026. It does not account for events that have not yet happened, and it is not investment advice.
3 Who finishes first in the first round — and why we part ways with the polls
| Poll average | Model expectation (median) | Chance of finishing first | Chance of winning outright, with no runoff | |
|---|---|---|---|---|
| Lula | 44.3% | 43.7% | 36% | 7% |
| Flávio Bolsonaro | 42.1% | 46.3% | 64% | 18% |
Lula leads the poll average. The model takes into account that in all 6 past elections PT's first-round gap to its rivals came out worse than in the polls, and shifts the gap by 4.8 points against PT on average. That is why, according to the model, Flávio Bolsonaro is more likely to finish first. The Kalshi market and the Plano Político model put Lula's chance of finishing first at 74% and 70% respectively [1][2].
A check on election night (4 Oct 2026). The Lula-minus-Flávio margin is 2.2 points in the poll average and -2.6 points in the model's expectation. If the actual margin turns out no smaller than in the polls, the historical skew has not repeated this time, and our estimate for Lula is too low.
The chance that it is all settled in the first round is 26% (Flávio 18%, Lula 7%). Kalshi prices this outcome at 4% [3] — one of our larger disagreements with the market: the model allows for large polling errors and therefore sees extreme outcomes more often.
First-round polls: weekly average, % of valid votes
| Date | Flávio Bolsonaro | Lula |
|---|---|---|
| 28 Aug | 41.1% | 44.5% |
| 4 Sep | 35% | 41.9% |
| 11 Sep | 40.5% | 44.4% |
| 18 Sep | 40.7% | 43.9% |
| 25 Sep | 40.9% | 43.4% |
| 2 Oct | 42.1% | 44.4% |
4 Scenarios: “if X happens, Lula's chance is Y”
We do not estimate the probability of the events themselves — only what happens to the chances if an event occurs. The size of each shift is taken from a real, measured past case.
| If… | Lula's chance |
|---|---|
| Our range | 20% to 29% |
| The polls miss as they did in the “Bolsonaro versus PT” elections (2018, 2022) | 19% |
| The polls miss as they did in the years when PT was in power (2006, 2010, 2014) | 13% |
| A repeat of 2006 — the only swing back to PT between the rounds | 50% |
| This time the head-to-head polls do not overstate Lula (no historical correction) | 41% |
| A scandal on the scale of May 2026 — against Flávio | 62% |
| The same scandal — against Lula | 8% |
Why the “polls miss as they did…” rows sit below our range. In those rows the polling error is taken exactly as it was in those years, with no softening and in both rounds at once. The main estimate allows for the error to be smaller this time, or even to run the other way, as in 2006. If history repeats itself in full, Lula's chance would be roughly 19%.
Where the size of the scandal comes from. After audio recordings of Flávio with the banker Vorcaro were published (13 May 2026), Lula gained an average of 3.4 points in head-to-head polls across the 6 pollsters with a poll both before and after [4]. Over the same period Flávio's support fell most visibly among evangelicals [5]. In the calculation the shift is applied to both rounds — to the first-round gap and to the runoff head-to-head polls — which is why the chance moves so much.
What it would take for the chances to even out. This time PT would have to fall short of its polls by noticeably less than usual. The first-round miss (result minus polls on PT's margin over its rivals) would have to be no lower than -0.5 points — this happened in 0 of 6 cases. The miss of the runoff head-to-head polls would have to be no lower than -0.1 points — this happened in 1 of 5 cases.
5 What moved the polls sharply
We found every sharp shift in the 2002–2026 polls (within a single pollster, over 21 days or less) and checked press reports and studies for what lay behind each one.
- A major scandal backed by direct evidence. May 2026: after the Vorcaro audio, Flávio lost 5.6 points in first-round polling (at 3 pollsters) and Lula gained 5.3 points (at 2) [4]. In 2006 the “dossier” scandal in the final weeks cost Lula a first-round victory [6].
- A little-known candidate's first week of mass television exposure. Augusto Cury in late August 2026 — up 7 points at 7 pollsters (the first debates, interviews on the main news programmes and the start of free electoral airtime, all in one week) [7][8]. In 2002, after the airtime began and José Serra went on the attack, Ciro Gomes collapsed — down 7 points [9][10]; in 2014, under attack in PT's advertising, Marina Silva lost 4.6 points [11][12].
- A change of candidate. In 2018 Fernando Haddad rose by 11.5 points once voters learned that he was Lula's candidate [13].
- In September 2026 Cury's votes began drifting back to the front-runners, and Flávio gained again — 3.9 points (at 2 pollsters), including among women and in the South [14][15].
- Did not move the gap in 2026: the increase in Bolsa Família payments [16], US tariffs [17], Flávio's meeting with Trump [18][19], Pablo Marçal's withdrawal [20].
The last first-round debate, on TV Globo (1 Oct 2026), did not take place. Lula had said in advance that he would not come [21]; the electoral court (TSE), at the request of his coalition, barred an empty lectern bearing his name and questions addressed to it, after which Flávio also pulled out shortly before air time and the network cancelled the programme [22]. Flávio immediately urged supporters to settle the election in the first round [22]. Whether any of this moved votes will show only in the polls of the last days before the vote [23]. The historical parallel is 2006: the “dossier” and Lula's refusal to debate cost him a first-round win, but in the runoff those votes came back [6].
6 Between the rounds, and a rule for election night
PT candidate's share of the two-candidate vote, % of valid votes (polls: [24][25][26][27]; results: [28][29][30][31][32][33]):
| Election | Head-to-head polls before the first round | First polls after the first round | Last polls before the runoff | Runoff result |
|---|---|---|---|---|
| 2002 | — | 65.6% | 63.9% | 61.3% |
| 2006 | 55.4% | 53.8% | 60.8% | 60.8% |
| 2010 | 60.3% | 53.9% | 56.5% | 56.1% |
| 2014 | 56.6% | 46.5% | 50% | 51.6% |
| 2018 | 49.6% | 42.4% | 44.2% | 44.9% |
| 2022 | 57.9% | 53% | 50.9% | 50.9% |
The first polls after the first round were worse for PT than the head-to-head polls before it in 5 of 5 cases — by between 1.7 points and 10.1 points. Voters of eliminated candidates went mostly to PT's opponents, even when their own candidate endorsed PT [34][35][36]. From the first polls after the first round to the runoff result, PT's share changed by between -4.3 points and 7.1 points, and it rose in 4 of 6 cases (in 3 of them the “first polls” were a single poll).
The rule for election night. Once the first-round result is known, our estimate of who wins the runoff goes out of date: the first Lula–Flávio head-to-head polls taken after the first round take its place. As a guide: historically those first polls started out worse for PT than the polls before the first round, and the last polls on the eve of the runoff were off by about 1.2 points on average (across 6 elections).
7 How we calculated
- The 2026 polls. We collected every available national poll; the calculation uses the average of the latest polls from 11 pollsters over the week before the cut-off. Polls from recent weeks were checked against the pollsters' own primary documents; the list of registered polls comes from the registry of the Superior Electoral Court (TSE) [23].
- Lessons from past elections. For 6 elections we compared the polls taken a few days before the vote with the official TSE results [28][29][30][31][32][33]; the polls come from the Pindograma database (data licensed with attribution) [37] and from Wikipedia tables [24][25][26][27]. The first-round miss (result minus polls on PT's margin over its rivals; a minus means PT's margin came out worse than in the polls):
| Year | What was compared | Result minus polls (+ = PT's margin beat the polls) |
|---|---|---|
| 2002 | First round, PT margin over rivals | -3.2 points |
| 2006 | First round, PT margin over rivals | -10.2 points |
| 2010 | First round, PT margin over rivals | -15.2 points |
| 2014 | First round, PT margin over rivals | -8.6 points |
| 2018 | First round, PT margin over rivals | -3.2 points |
| 2022 | First round, PT margin over rivals | -2.6 points |
- The simulation. We play out the first round 200,000 times with a polling error like those of the past, determine the runoff pair, and play out the runoff from that pair's polls with the historical correction.
- Testing on the past. We ran the model on each past election without showing it the result. The first-round correction cut the polling miss from 8.6 points to 6.9 points on average — this rule has been tested by history. History cannot test the runoff correction: the miss with it is 5.5 points, without it 5.4 points, there are 5 cases, and a single year decides the answer. In picking the winner, the test showed no advantage over a simple poll average: there are few past elections, and PT won almost all of them.
The chart below shows how Lula's chance changes under other reasonable assumptions. Some rows go beyond our range: its edges are a guide, not hard limits.
Lula's chance under other reasonable model assumptions
| Assumption | Low | High |
|---|---|---|
| Both corrections, not softened | 19% | 29% |
| Runoff correction, not softened | 20% | 29% |
| Stronger first-round correction | 24% | 29% |
| Round link: simple regression | 25% | 29% |
| Strong link between rounds | 29% | 33% |
| Runoff correction softened less | 26% | 29% |
| Minor votes by second preference | 26% | 29% |
| No prior assumptions on spread | 29% | 31% |
| Mean poll miss, not median | 28% | 29% |
8 Our model versus betting markets and other forecasts
| Who | Flávio's chance | Lula's chance | Lula first in the first round |
|---|---|---|---|
| Our model | 71% to 80% | 20% to 29% | 36% |
| Polymarket (trading volume $161 million) [38] | 56% | 44% | — |
| Kalshi [39][1] | 57% | 43% | 74% |
| PredictIt [40] | 61% | 45% | — |
| Plano Político model [2] | 58% | 42% | 70% |
| ÍRIS model [41] | 68% | 32% | — |
| PollingData model (BBC Brasil partner) [42] | — | — | 79% |
In our model the favourite is Flávio Bolsonaro; on Polymarket, the largest market, the favourite is Flávio Bolsonaro. Our estimate for Lula is below the market's, and on the question of who finishes first we part ways with the markets and with other models. This does not prove the market wrong: our first-round forecast rests on a correction that history has confirmed (see “How we calculated”), while the headline chance depends on the runoff correction, which history cannot test. A betting-market price is not an exact probability either: it is shaped by fees, traders' tendency to overpay for long shots (the favourite–longshot bias) and the size of the market.
9 What the model does not know
- The runoff correction is the key assumption. In 4 of 5 past cases, head-to-head polls taken before the first round overstated PT (by 4.8 points in 2018 and by 7 points in 2022), while in 2006 they understated it — by 5.4 points: that year, voters who had punished Lula in the first round came back to him in the second [6]. We carry over the median of these misses but soften it, because there are so few cases. How much it is softened is set, in effect, by 2006 alone: that one year greatly widens the estimated spread. Without the softening, Lula's chance is 20% (the bottom of the range); with no correction at all, it is 41%. A single election moves the answer a lot: without 2006, Lula's chance is 15%; without 2014, it is 38%.
- Survey mode. In 2026 more pollsters interview by phone and online (7, against 3 face-to-face; for 1 the method is unknown). The face-to-face minus phone-and-online gap in Lula's share is 2.2 points (from 0.5 points to 3.9 points). The past picture is mixed (result minus polls; a minus means the polls overstated PT):
| Election | First round: face-to-face | First round: phone and online | Head-to-head polls: face-to-face | Head-to-head polls: phone and online |
|---|---|---|---|---|
| 2018 | -7.1 points | -0.5 points | -6.5 points | -4.5 points |
| 2022 | -8.5 points | -4 points (excluding one outlier pollster; with it, 0.3 points) | -8.1 points | -6.7 points |
In both the 2018 and the 2022 first rounds, phone and online pollsters overstated PT less than face-to-face ones; in head-to-head polls the gap between them is smaller. If today's mix of pollsters does overstate PT less than in the past, it is more likely to do so in the first round — in which case Lula's chance is higher than our estimate. With only 6 elections, there is no way to measure this.
- Little history. We have 6 elections with both results and polls; for the runoff, 5. Any rule drawn from so few cases remains rough.
- Health. Lula is 80; by the IBGE life table, the probability that he does not survive to the runoff is about 0.4% [43]. This is a lower bound on the risk of his dropping out: the table does not account for illness or withdrawal from the race.
10 Power players and social media
Neither feeds into our probabilities: the players' positions affect votes through the polls, which we have already taken into account.
Power players. We collected 69 verified positions, with quotes and copies of the sources. The highlights:
| Who | Which side |
|---|---|
| Governors of 7 states, including São Paulo and Minas Gerais (40.4% of the country's voters) [44][45][46] | Flávio |
| Governors of 8 states, mostly in the Northeast (22.5% of voters) [47][48] | Lula |
| Governors of 4 states (17.3% of voters): Paraná and Rio Grande do Sul back Caiado; Espírito Santo and Pernambuco have not picked a side [49][50][51][52] | — |
| Evangelical leaders: Pastor Silas Malafaia and the Renascer church [53][54]; some progressive pastors back Lula [55] | mostly Flávio |
| Centrist parties MDB, União Brasil, Republicanos and PRD-Solidariedade — neutral in the first round; the head of the PP is personally voting for Flávio [56][57][58][59][60] | neutral |
| The CUT trade-union federation and the Landless Workers' Movement (MST) [61][62] | Lula |
| Financial markets: the stock market rises on news favouring Flávio and falls on scandals around him [63][64][65] | mostly Flávio |
| Foreign leaders: Trump has spoken warmly of Flávio, and Milei came to show his support [66][67]; Xi Jinping, in a conversation with Lula, backed Brazil against “external interference” [68] | divided |
Social media: a radar for events, not votes. Over the past month we collected 2,132 posts by Lula and Flávio on 5 platforms. The most prominent: Lula's post on a decree banning online betting and casinos (26 Sep 2026, 1,503,288 interactions) [69]; Flávio's reply, “Lula will bet the election on ‘Tigrinho’”, an online casino game (25 Sep 2026, 917,515 interactions) [70]; and his post “Lula is not fit to be president” (29 Sep 2026, 924,229 interactions) [71]. Why this is not the same as votes: in 2022 Quaest's digital popularity index between the rounds (7 Oct 2022) put Jair Bolsonaro ahead of Lula (88.1 to 83.7) [72], yet Lula won the runoff with 50.9% [33]. In 2026 the indices contradict each other: Nexus has Lula ahead [73], Datrix has Flávio [74].
11 Calendar: what to watch
- 1 Oct 2026 — the last first-round debate, on TV Globo, was cancelled: Lula declined in advance, Flávio pulled out of it after the TSE ruling [21][22].
- 3 Oct 2026 — the final wave of polls (Datafolha, Quaest, AtlasIntel and others) [23]. We will update the forecast on the morning of election day.
- 4 Oct 2026 — first round. In the evening: the result, and the election-night rule (see above).
- 11 Oct 2026 — runoff debate on Record [21].
- 25 Oct 2026 — runoff [75].
12 What went into the research, and how to check it yourself
| What went in | Count |
|---|---|
| 2026 polls analysed | 127 |
| polling firms | 19 |
| recent polls checked against the pollster's own document | 58 |
| polls from past elections 2002–2022 | 396 |
| past elections used to test the model | 6 |
| sources with a saved copy | 334 |
| documents downloaded and parsed | 3,778 |
| candidate posts on social media | 2,132 |
| social-media platforms | 5 |
| verified positions of key players | 69 |
| governors on the power map | 19 |
| campaign donation records (TSE) | 3,475 |
| betting-market contracts | 61 |
| days of Polymarket price history | 379 |
| assumption variants in the sensitivity check | 39 |
| “what if” scenarios | 7 |
| model simulation runs | 8,600,000 |
| automated quality checks | 19 |
How to check it yourself.
- Open any footnote: every source has a link, an access date and a saved copy.
- Check the poll averages against the Ranking dos Políticos and Plano Político aggregators [76][2]; a gap of a fraction of a point is normal.
- Check market prices on Polymarket and Kalshi [38][39]: they change by the hour.
- Past election results are in the TSE's open data [33].
| No. | Source | Date accessed | Copy |
|---|---|---|---|
| 1 | kalshi.com link | 2 Oct 2026 | copy |
| 2 | planopolitico.com.br link | 2 Oct 2026 | copy |
| 3 | kalshi.com link | 2 Oct 2026 | copy |
| 4 | mais.opovo.com.br link | 1 Oct 2026 | copy |
| 5 | www.portaltela.com link | 1 Oct 2026 | copy |
| 6 | biblat.unam.mx link | 1 Oct 2026 | copy |
| 7 | veja.abril.com.br link | 1 Oct 2026 | copy |
| 8 | valor.globo.com link | 1 Oct 2026 | copy |
| 9 | www.redalyc.org link | 1 Oct 2026 | copy |
| 10 | www.redalyc.org link | 1 Oct 2026 | copy |
| 11 | www.em.com.br link | 1 Oct 2026 | copy |
| 12 | ihu.unisinos.br link | 1 Oct 2026 | copy |
| 13 | www.bemparana.com.br link | 1 Oct 2026 | copy |
| 14 | g1.globo.com link | 1 Oct 2026 | copy |
| 15 | media.folha.uol.com.br link | 1 Oct 2026 | copy |
| 16 | g1.globo.com link | 1 Oct 2026 | copy |
| 17 | www.opovo.com.br link | 1 Oct 2026 | copy |
| 18 | www.poder360.com.br link | 1 Oct 2026 | copy |
| 19 | www.opovo.com.br link | 1 Oct 2026 | copy |
| 20 | g1.globo.com link | 1 Oct 2026 | copy |
| 21 | www.correiobraziliense.com.br link | 1 Oct 2026 | copy |
| 22 | www.otempo.com.br link | 2 Oct 2026 | copy |
| 23 | cdn.tse.jus.br link | 1 Oct 2026 | copy |
| 24 | en.wikipedia.org link | 1 Oct 2026 | copy |
| 25 | en.wikipedia.org link | 1 Oct 2026 | copy |
| 26 | en.wikipedia.org link | 1 Oct 2026 | copy |
| 27 | en.wikipedia.org link | 1 Oct 2026 | copy |
| 28 | dadosabertos.tse.jus.br link | 1 Oct 2026 | copy |
| 29 | dadosabertos.tse.jus.br link | 1 Oct 2026 | copy |
| 30 | dadosabertos.tse.jus.br link | 1 Oct 2026 | copy |
| 31 | dadosabertos.tse.jus.br link | 1 Oct 2026 | copy |
| 32 | dadosabertos.tse.jus.br link | 1 Oct 2026 | copy |
| 33 | dadosabertos.tse.jus.br link | 1 Oct 2026 | copy |
| 34 | sul21.com.br link | 1 Oct 2026 | copy |
| 35 | ihu.unisinos.br link | 1 Oct 2026 | copy |
| 36 | www.cartacapital.com.br link | 1 Oct 2026 | copy |
| 37 | github.com link | 1 Oct 2026 | copy |
| 38 | polymarket.com link | 2 Oct 2026 | copy |
| 39 | kalshi.com link | 2 Oct 2026 | copy |
| 40 | www.predictit.org link | 2 Oct 2026 | copy |
| 41 | voga-iris.com link | 2 Oct 2026 | copy |
| 42 | flex.pollingdata.com.br link | 2 Oct 2026 | copy |
| 43 | ftp.ibge.gov.br link | 1 Oct 2026 | copy |
| 44 | pt.wikipedia.org link | 1 Oct 2026 | copy |
| 45 | www.otempo.com.br link | 1 Oct 2026 | copy |
| 46 | www.gazetadopovo.com.br link | 1 Oct 2026 | copy |
| 47 | pt.wikipedia.org link | 1 Oct 2026 | copy |
| 48 | www.infomoney.com.br link | 1 Oct 2026 | copy |
| 49 | www.gazetadopovo.com.br link | 1 Oct 2026 | copy |
| 50 | sbtnews.sbt.com.br link | 1 Oct 2026 | copy |
| 51 | www.folhavitoria.com.br link | 1 Oct 2026 | copy |
| 52 | www.diariodepernambuco.com.br link | 1 Oct 2026 | copy |
| 53 | portaldeprefeitura.com.br link | 1 Oct 2026 | copy |
| 54 | viva.com.br link | 1 Oct 2026 | copy |
| 55 | www.diariodocentrodomundo.com.br link | 1 Oct 2026 | copy |
| 56 | www.poder360.com.br link | 1 Oct 2026 | copy |
| 57 | www.poder360.com.br link | 1 Oct 2026 | copy |
| 58 | www.correiodamanha.com.br link | 1 Oct 2026 | copy |
| 59 | www.congressoemfoco.com.br link | 1 Oct 2026 | copy |
| 60 | revistaoeste.com link | 1 Oct 2026 | copy |
| 61 | www.cut.org.br link | 1 Oct 2026 | copy |
| 62 | www.gazetadopovo.com.br link | 1 Oct 2026 | copy |
| 63 | www.bloomberglinea.com.br link | 1 Oct 2026 | copy |
| 64 | www.correiodamanha.com.br link | 1 Oct 2026 | copy |
| 65 | www.dgabc.com.br link | 1 Oct 2026 | copy |
| 66 | www.congressoemfoco.com.br link | 1 Oct 2026 | copy |
| 67 | www.band.com.br link | 1 Oct 2026 | copy |
| 68 | english.ahram.org.eg link | 1 Oct 2026 | copy |
| 69 | www.instagram.com link | 1 Oct 2026 | copy |
| 70 | www.instagram.com link | 1 Oct 2026 | copy |
| 71 | www.instagram.com link | 1 Oct 2026 | copy |
| 72 | www.opovo.com.br link | 1 Oct 2026 | copy |
| 73 | istoedinheiro.com.br link | 1 Oct 2026 | copy |
| 74 | www.jornalopcao.com.br link | 1 Oct 2026 | copy |
| 75 | pt.wikipedia.org link | 1 Oct 2026 | copy |
| 76 | ranking.org.br link | 1 Oct 2026 | copy |
All data come from public sources. This is an estimate of probabilities, not investment advice.