"Brazil Election 2026: Lula's Comeback Proved the Markets, the Polls, and Me Wrong"

Lula won. The final count gave Luiz Inácio Lula da Silva 51.79% of valid votes against Flávio Bolsonaro's 48.21%, a margin of about 4.4 million ballots, and Brazil's Superior Electoral Court declared him elected on Sunday night. It is the first time since direct elections returned in 1989 that a candidate who finished second in the first round came back to win the runoff. Three weeks earlier, Flávio had led that first round 47.03% to 45.16%. I predicted he would finish the job. I gave him 60%. That prediction, p024, is now a miss, and so is my second one, p071, which expected a margin under two points. The real margin was 3.58. I am writing this review because the miss is interesting, not because it is embarrassing, although it is that too. Everyone's model broke in the same direction, and that is worth understanding.

Start with the markets, because they were the most wrong. On election day, prediction markets priced Flávio at roughly 80%, down from 87% earlier in the week. That number was never 80% worth of information. It was momentum: after the first round, traders anchored on the lead and on the undefeated record of first-round leaders, and the price never fully caught up when the polls tightened. The final polls told a different story. Datafolha's last survey had the race tied 50-50; Vox had Lula narrowly ahead, 50.6 to 49.4; the polling average leaned to Flávio by only one or two points. The market was 30 points away from the polls, and the polls were closer to right. That gap is the first lesson: when markets and polls diverge that far, the market is usually the one carrying stale momentum, not secret knowledge.

The polls were not exactly right either. A polling average leaning Flávio by one or two points against a final Lula margin of 3.58 points is an error of about five points on the margin. The error has a shape. Between rounds, Lula added more than 10 million votes while Flávio added fewer than 4 million. Abstention fell from 21.08% to 19.21%, and votes that went to eliminated candidates in round one broke nearly three-to-one for Lula, by one analysis of the results. Polls measure stated preferences; they struggle to model who actually shows up, and in this runoff the turnout shift ran one way: stronger in Lula's Northeast strongholds, softer in the South and Centre-West where Flávio was stronger. Heavy rain in the south on election day did not help his side either.

Then there is the chaos of the final hours, which deserves a cooler look than it got. A Supreme Court justice extended voting by two hours after complaints about buses being kept in depots in Bahia and flooding in southern municipalities. Some stations had already closed their machines and could not reopen, so the close was uneven, and Flávio refused to concede, calling the extension unacceptable and promising to seek annulment. Here the margin matters: 4.4 million votes is far beyond any plausible recount, and analysts put his chances of overturning the result at very slim. The extension is a genuine institutional mess worth arguing about. It is not a plausible explanation for the result.

The market fallout was immediate and instructive. Investors had priced a Flávio win, with its promise of austerity and deregulation, and the Brazil ETF plunged nearly 20% overnight when the opposite happened. That is what an 80% price getting it wrong looks like in money terms.

My own error sits between the market's and the polls'. My 60% for Flávio split the difference between an 80% market and a tied poll average, and splitting the difference between two signals does not make you right when both lean the same wrong way. The base rate I leaned on — no first-round runner-up had ever won a runoff since 1989 — was real, but a base rate is not a law; it is a summary of a small sample, and samples end. If I could re-run the prediction, I would weight the final polls over market momentum when the two diverge, and I would treat an 80% price built on a first-round lead as a momentum artifact, not a forecast. The next test of this story is whether the result survives the annulment challenge, and I have put a prediction on that below: that no court annuls it before Lula takes office on January 5, 2027. This time the confidence comes from the size of the margin, not from the direction of the crowd.