AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
BrazilThe week of 31 August to 4 September compressed the presidential spread for the second time running, and the mechanics were the opposite of the previous week. The gap between Lula and Flávio Bolsonaro in the presidential contract closed from 16.20pp to 14.90pp, a compression of 1.30pp across five sessions, against 9.70pp in the week of 24 to 28 August. What changes is not the size, it is where it came from: in the previous week the leader gave up 5.00pp and the runner-up rose 4.70pp, both legs moving almost half and half. In this one the leader closed the five sessions exactly where he opened, at 55.50%, and the entire compression came from the runner-up, who rose from 39.30% to 40.60%. Two weeks of compression produced by different engines are not the same trend measured twice.
And the presidential book was the quietest part of a violent week. The repricing happened in the contracts around it. The contract on a Supreme Court justice leaving by impeachment before 2027 went from 3.60% to 15.10%, a rise of 11.50pp that multiplies the price by more than four in five sessions, along a path that does not look like convergence: 3.60, 11.50, 6.10, 12.20, 15.10. In the first-round third-place contract, Ronaldo Caiado lost 21.50pp, from 30.50% to 9.00%, while Renan Santos gained 5.00pp. A book of USD 1.01M reordered its field in a single week.
The six national polls in the window do not settle the reading, they widen it. The first-round gap they measure runs from 3 points, in PoderData of 3 September, to 9.7 points, in AtlasIntel of 31 August, with 6.7 points of dispersion between houses on the same question and the same week. And in the runoff two of the six do not put the leader ahead: PoderData measures the runner-up winning 45 to 44, and Real Time Big Data measures a 44 tie. The price, which closed the week paying 55.50% on the leader, is pricing a scenario that two of the six ballot readings do not support.
1.Executive Summary
The theme of the week is the separation between where the price moved and where the attention was. The presidential book, which holds USD 144.15M in accumulated volume, was the quietest of the five: the leader did not vary a single point between Monday's open and Friday's close. The movement showed up in the peripheral contracts, and it was large. The Supreme Court impeachment contract jumped 11.50pp on a base of 3.60%, and the first-round third-place book reversed the order between Ronaldo Caiado and Renan Santos with a 21.50pp move in the former. In both cases accumulated volume is small, USD 87,664 and USD 1.01M, and that is a structural feature of those books rather than noise to discount: a thin price moves further per order, which is exactly why book size sits beside the number.
2.Why AFOS does not smooth
The six national polls of the week asked the same question, with overlapping fieldwork and samples between 2,000 and 5,014. An arithmetic mean of them returns a clean number and erases the one piece of information the set carries, which is the size of the disagreement.
Why it matters: the 6.7-point range is larger than the declared margin of any of the six, so the disagreement is not sampling noise, it is a difference of method or of screen. Whoever trades the average trades a number that none of the six houses measured, and loses the information that the runoff reading is split down the middle among them.
Issue №15 opened by saying that the 9.70pp compression of the week of 24 to 28 August came split almost evenly between the two legs, and that this set it apart from the week before. The following week kept the direction and changed the engine: it compressed 1.30pp with the leader motionless. What is worth recording here is that the direction repeated and the mechanics did not, and a reading that had extrapolated the mechanics of №15 would have missed this week even while getting the sign right.
3.Weighted scenarios for the week
The eleven national polls registered with the TSE for release between 7 and 11 September concentrate, in a single week, more readings than the series has had in any previous window. The scenarios below concern what the pricing does when they arrive.
The eleven national polls arrive distributed around the central band of the previous week, with the first round between 5 and 8 points and the runoff inside the declared margin. In that case the presidential spread settles between 13pp and 17pp without breaking any recent boundary, and the impeachment contract gives back part of the rise, because the 11.50pp jump was built with no new institutional act on three of the five sessions. The third-place book tends to consolidate Augusto Cury above 50% with Renan Santos in the 20% to 27% range, and Ronaldo Caiado below 12%, keeping the order the week produced.
The two houses that measured the runner-up ahead or level in the runoff were not the exception, and the reading repeats in two or more of the eleven. The Veritá poll released on 6 September, with fieldwork from 1 to 4 and 3,804 interviews, already put the runner-up numerically ahead in the first round, at 38.9% against 38.4%, a distance that fits inside the margin of 2 points and therefore measures a statistical tie. If that pattern shows up in houses with higher reliability, the presidential spread breaks below 13pp and the price has to choose between the ballot and its own series, which has not recorded the leader below 54.50% since the start of August.
A new institutional act in the Supreme Court case puts the impeachment contract back above Friday's 15.10% and keeps it there. The week produced two acts on 5 September, a proposed amendment to the court's internal rules filed by Gilmar Mendes and a request for accountability over a breach of tax secrecy made by Cristiano Zanin, and the price reacted to neither: it closed Friday at 15.10% and stood at 8.60% on the confirmed reading of 6 September. A book that rises 11.50pp in a week and gives back 6.50pp over the following weekend is not pricing a path, it is testing a range.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| Lula, presidential | 55.50% | 0.00pp | USD 9,829,870 | Closed the five sessions exactly where he opened. The intra-week path had a single move, the 1.00pp fall on Thursday followed by full recovery on Friday. It is the first week in the series in which the leader returns zero variation and the spread compresses anyway. |
| Flávio Bolsonaro, presidential | 40.60% | +1.30pp | USD 9,743,848 | The whole of the week's compression. Rose from 39.30% to 44.00% on Thursday and gave back 3.40pp on Friday, ending 1.30pp above the open. His accumulated volume sits USD 86 thousand from the leader's, effectively level in a USD 144.15M book. |
| Supreme Court impeachment before 2027 | 15.10% | +11.50pp | USD 87,664 | Largest relative move in the grid, with the price multiplied by more than four. The path was 3.60, 11.50, 6.10, 12.20 and 15.10, that is two double-digit rises separated by a 5.40pp give-back mid-week. Book of USD 87,664 accumulated, the smallest in the grid. |
| Caiado, first-round third place | 9.00% | -21.50pp | USD 114,472 | Largest absolute fall in the grid. Went from 30.50% on Monday to 14.00% on Tuesday and 8.50% on Wednesday, and stayed in that range. Lost second position in the book to Renan Santos within the same week. |
| Renan Santos, first-round third place | 23.50% | +5.00pp | USD 278,938 | Rose to 30.00% on Tuesday, gave back to 22.00% on Thursday and closed at 23.50%. Took second position in the third-place book without the leader of that book moving comparably. |
| Augusto Cury, first-round third place | 53.20% | +0.80pp | USD 125,708 | The book's leader crossed the week almost flat in net terms, with an internal amplitude of 12.50pp between the 46.50% low on Tuesday and the 59.00% high on Thursday. The reordering behind him did not pass through his price. |
| Flávio Bolsonaro, first-round runner-up | 88.50% | +2.00pp | USD 612,294 | Closed the week at 88.50% after pulling back to 84.00% on Thursday. This is the contract where the market concentrates its conviction about that name, and it pays more than twice what the presidential contract pays. |
| PL, Senate majority | 82.50% | +3.50pp | USD 260,543 | Rose from 79.00% to 83.50% on Wednesday and stabilised. The MDB, the book's runner-up, gave up 2.20pp over the same period, in a contract with USD 8,909 accumulated where a small order shifts several points. |
| Annual inflation between 5.00% and 5.49% | 42.35% | 0.00pp | USD 15,649 | The modal band of the inflation book did not move on the week. It is the only contract in the grid whose price does not respond to the electoral calendar, and it serves here as a baseline for separating market movement from theme movement. |
5.Liquidity and market structure
The two largest accumulated volumes in the book pay 0.05% and 1.75%. Tarcisio de Freitas, with USD 14.07M, is not a presidential candidate and is running for re-election in São Paulo; the contract remains open by inheritance from the market's design. Renan Santos, with USD 12.72M, is the second largest volume in the book and the fourth price among the names above 1%. Accumulated volume is money that has already passed through the contract since it opened and says nothing about which way it went, so the correct reading is that these two lines concentrated trading at some point in the cycle, not that the market considers them likely today.
Volume is accumulated since each contract opened, measured on the reading of 6 September at 16:53 UTC. The AFOS does not cite book depth in editorial text: low liquidity in a continuously arbitraged market does not mean the price is wrong, and exposing the technical number invites the opposite reading.
6.Calendar of price-relevant prints
| Date | Sample | Why it matters | |
|---|---|---|---|
| Sep 7 | Quaest/Genial | n=2,004 | Tier 1. Measured 7 points in the first round on 2 September and 42 to 41 in the runoff, the tightest reading among the reliability-5 houses of the week. |
| Sep 8 | Nexus | n=2,000 | The BTG/Nexus series measured 6 points on 31 August. A repeat of that level confirms the central band; a break below 4 moves the series closer to PoderData. |
| Sep 9 | PALVER | n=5,000 | Second largest sample of the week and a method distinct from the others. A large sample narrows the declared margin and makes the reading harder to dismiss as an isolated point. |
| Sep 10 | AtlasIntel/Bloomberg | n=5,000 | Tier 1. This was the house that measured the widest gap of the previous week, 9.7 points with a margin of 1. Confirmation keeps the upper edge of the dispersion where it is. |
| Sep 11 | Veritá | n=40,500 | The largest sample ever registered in the series for a national poll, eight times the second largest in the window. The house carries a reliability of 2 on the AFOS scale and its previous round put the runner-up ahead. |
| Sep 11 | Datafolha | n=2,002 | Tier 1. Measured 7 points in the first round and 46 to 44 in the runoff on 3 September. It is the longest series in the window and the reference against which the others are read. |
Eleven national polls are scheduled for release between 7 and 11 September, and the table lists the six with the greatest weight by sample or by reliability. Registration with the TSE is a commitment to publish, not a guarantee: institutes delay and cancel, and the scheduled date anticipates no number. A method caveat on the Real Time Big Data entry, which does not appear in this table: the national scope of that registration is derived by us from the text of its sampling plan, and the house check does not sustain the label, because that pollster's sampling plan declares a national universe in its state-level polls too.
7.Watch list, week triggers
- The leader breaking below 54.50% The presidential contract has not recorded the leader below that level since the start of August. With eleven national polls arriving across five days, it is the first testable boundary, and a break below would put the price under the band it held throughout the two-week compression.
- The second house measuring an inversion in the first round The Veritá poll of 6 September put the runner-up ahead by 0.5 point, inside the margin, and it is the second national poll of the window in that order after Gerp of 26 August. A third, from a house with reliability 4 or 5, stops being an isolated point.
- The impeachment contract above 15.10% The week's high became the reference level. The price has already given back 6.50pp since Friday and stands at 8.60% on the confirmed reading of 6 September. Returning above the high with a new institutional act is different from returning without one, and the distinction defines whether the book is pricing a path or testing a range.
- The order of the third-place book Renan Santos passed Ronaldo Caiado on a 21.50pp move in the latter. The book holds USD 1.01M accumulated and reordered in three sessions, so the stability of the new order is the question, not the order itself.
- Whether the dispersion of the six houses narrows The 6.7-point range between AtlasIntel and PoderData is wider than the declared margin of any of them. If the eleven polls of next week deliver similar dispersion, the disagreement is structural to method rather than to sample, and that changes how the set should be read from here on.
8.Methodology
The AFOS does not produce a weighted average and does not arbitrate between sources. The reading crosses three signals that measure different things: prediction market price, which is implied probability traded in dollars; polls registered with the TSE, which are declared voting intention; and press flow on institutional acts. Where they diverge, the divergence is the data, and reducing it to a single number would erase exactly the information that makes it useful.
Prices in this issue come from the confirmed reading of 6 September at 16:53 UTC, and the AFOS only publishes a price that two independent readings, taken eight minutes apart, confirm within 0.20 point. Weekly changes are measured close to close, from the last point of 31 August to the last point of 4 September, over the complete series recorded since 14 April. Every superlative declares the window in which it holds. Code and method are open under Apache 2.0 at github.com/andrefelipe-afos/afos-analitica-2026.
9.Additional reading · macro coverage
Complementary reading for macro context. The AFOS is a primary source on the divergence between market and polls; the links below cover the fiscal and political backdrop the contracts do not measure directly.
- Valor Econômico · Daily coverage of the 2026 election and the fiscal impact of the electoral calendar.(paywall), valor.globo.com/politica/eleicoes-2026/
- Reuters Brazil · English-language flow on politics and currency, useful for the foreign investor's reading., www.reuters.com/world/americas/
- Bloomberg Línea Brasil · Market analysis focused on Brazilian assets and the political risk premium.(paywall), www.bloomberglinea.com.br/
- Poder360 · Tracking of registered polls and institutional agenda, open access., www.poder360.com.br/poder-eleicoes-2026/
- TSE public register · Official search for registered electoral polls, the primary source of this issue's calendar., divulgacandcontas.tse.jus.br/divulga/
Get it by email
AFOS Tradeoff is the weekly read on Brazilian political risk, for people who decide on data rather than headlines.
No spam. Unsubscribe anytime.
After signing up you choose whether to receive it in English, Portuguese or Spanish.