The Polyvox Read — Brazil, 20 days out

Edition 1. Data as of September 6, 2026, from the Polyvox model.

The runoff is a coin toss. The kingmaker is not.

Brazil votes on October 4. Every major pollster has the likely second round — Lula da Silva against Flávio Bolsonaro — inside the margin of error: Datafolha 46–44, PoderData 44–45, Quaest 40–42. Our model puts the government’s chance of holding the presidency at 47%. We label that what it is: a toss-up, low confidence.

That number is not the interesting part. What matters for anyone with money or operations in Brazil is why the race is tied, and which two events between now and Round 1 will move it.

What happened

The incumbent is running on an approval rating of roughly 44–45% at our September 6 cut, down about three points over the quarter, against 51% disapproval; Datafolha’s September 11 read has it recovering to 47% approve, 50% disapprove. In our data on 34 Latin American elections since 2004, that is still the danger zone: high enough to keep the opposition from consolidating an anti-incumbent wave, too low, and too recently falling, to carry a reelection on its own.

The news cycle is doing the rest. Public debt crossed 82.5% of GDP and state-company deficits hit a semester record, both framed by mainstream outlets as governance failures. The federal police inquiries involving the president’s son stayed at the top of the agenda through Lula’s Jornal Nacional interview. Our sentiment index for the Aug 30–Sep 6 week reads −60, the most negative of the cycle. The one positive beat, the Sep 6 Independence Day address, landed with the government’s own audience and nobody else.

On the other side, Flávio Bolsonaro carries his father’s endorsement from house arrest and the international right’s, and the amnesty question is his platform’s center of gravity. Structurally, this is Peru 2011: the heir of a jailed ex-president reaching a runoff against an incumbent-side candidate, with a centrist bloc deciding the outcome.

Why it is tied: one bloc, eleven points

Augusto Cury, running for Avante, holds about 11% of first-round intention with the lowest rejection rate in the field (28%). That bloc is the single highest-variance variable in our model, and its direction is unresolved.

If it breaks 60% or more to Flávio, as anti-incumbent protest votes did in comparable runoffs, the race tips. If Cury endorses Lula between rounds, or if the amnesty platform pushes his voters toward a “democratic guardrails” vote, the incumbent holds. Neither outcome is priced in yet, by us or by the polls.

The two dates that matter

September 16, Copom. A rate decision read as further tightening reopens the fiscal-deterioration frame in the final fortnight. A hold with an IPCA-deceleration story does the opposite: it is the most plausible way the government’s sentiment score recovers above −30, which is the threshold our incumbent-retention rules failed on this week.

September 14–18, Datafolha. The last major read before Round 1. If the second-round simulation moves outside the margin in either direction, the toss-up label goes with it.

What it means if you operate in Brazil

Plan for a 50/50 through October 25, not for a favorite. In practice: any regulatory or fiscal exposure that depends on which coalition holds the Planalto in 2027 should be hedged now, not after Round 1. Assume that whoever wins inherits a debt-to-GDP narrative that forces a post-election fiscal correction; the difference between the candidates is who pays for it, not whether it happens. And watch the Cury bloc’s polling in the week after the 16th more closely than the headline numbers.

On the record

As of September 6: government retention 47%, toss-up, low confidence. Projected runoff Lula vs. Flávio Bolsonaro. Decisive variable: the direction of the Cury bloc. Next update after Datafolha.

We will publish the result against this call after October 25.