AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
BrazilThe week of August 24 to 28 recorded the largest weekly compression of the presidential spread since the series began. The gap between Lula and Flávio Bolsonaro closed from 29.60pp to 19.90pp in the presidential contract, a compression of 9.70pp across five sessions, against 3.80pp the week before and a median of 1.40pp among the weeks that compressed. Measured over the 17 Mon-Fri weeks of the series, which begins on April 20, no other came close.
The compression came split almost evenly, and that sets it apart from the previous week. The leader gave up 5.00pp, from 62.50% to 57.50%, and the runner-up rose 4.70pp, from 32.90% to 37.60%. In August 17-21 the move had come more from the runner-up than from the leader; this time both sides contributed in similar proportion. The gap closed lower in all five sessions, without a single reversal: 29.60pp, 27.30pp, 26.90pp, 23.00pp and 19.90pp.
The choice of endpoint does not change this week’s reading. Monday had a single capture in the presidential contract, so close, open and intraday extreme are the same point, and all three criteria return 9.70pp. This issue uses the daily close across every contract, as issue №14 did, and the criterion is stated in the methodology.
The declared ballot did not follow the price with the same single direction. The four national polls published in the window disagree on the sign of the runoff: Gerp on August 26 gives Flávio Bolsonaro 47% against 42%, and Indexa/Broadcast, published on the same day, gives Lula 46% against 41%. That is ten points of spread between two pollsters with overlapping fieldwork, and the simple average of the four returns a 0.5pp lead for the leader, a value none of them measured.
1.Executive Summary
The theme of the week is speed. The presidential spread compressed in four consecutive weeks, and the magnitude rose in three of them: 1.40pp between August 3 and 7, 0.80pp between the 10th and the 14th, 3.80pp between the 17th and the 21st and 9.70pp in this one. Across the 17 Mon-Fri weeks of the series, which begins on April 20, ten widened the spread and seven compressed it, and the largest previous compression was precisely that of August 17-21. A second contract recorded an even larger move in absolute terms: in first-round third place, Renan Santos gave up 18.50pp and Augusto Cury rose 25.00pp in the same window, ending Friday tied with Ronaldo Caiado at 35.50%. Both moves occur while the four national polls of the week disagree on who leads the runoff, with ten points separating the extremes.
2.Why AFOS does not smooth
The four national polls published between August 24 and 28 tested the same runoff matchup. Aggregating them into an average produces a number that describes the set and none of the measurements, and erases exactly what makes the week informative.
<strong>Why it matters:</strong> Gerp and Indexa/Broadcast published on the same August 26, with overlapping fieldwork, and arrived at opposite signs ten points apart. The disagreement repeats in the first round, with a smaller spread and the same shape: Gerp is the only one of the four that puts the runner-up ahead, at 38% against 37%, while Indexa gives 39% to 34% and BTG/Nexus 41% to 37%. That is six points between the extremes, and the median of the four sits at a three-point lead for the leader. The average of the four in the runoff returns half a point and suggests stability. What exists is disagreement among pollsters over who is ahead, and it is that disagreement the price has been pricing toward the runner-up for four weeks.
The same framework applied to Brazil, measuring the distance between the prediction market and the polls and checking the signal against the actual election result, has been tested against eleven elections on four continents, among them India, France, the United Kingdom, Mexico and the United States (2024), South Korea, Chile, Germany and Canada (2025) and Peru and Colombia (2026). India is the only case in which market and polls erred together in the same direction. The method reports convergence and divergence with no retroactive adjustment. The dataset is open under Apache 2.0 and carries a permanent DOI at the Harvard Dataverse.
3.Weighted scenarios for the week
The TSE registry lists eight national polls scheduled for release between August 31 and September 3, three of them on Monday. It is the largest block of readings in the series across four days, and it is what resolves the scenarios below.
The eight scheduled national polls arrive distributed around the BTG/Nexus and PoderData range, with the runoff inside the margin and the first round between three and five points, and Gerp stays isolated as the extreme point of the window. In that case the compression loses traction around 14pp to 18pp, because the spread already embeds a reading more favourable to the runner-up than the median of the four polls registered during the week. The third-place contract tends to settle the tie between Ronaldo Caiado and Augusto Cury around 35% to 43%, with Renan Santos below both.
The national polls confirm the Gerp reading and the runoff starts showing the runner-up ahead in more than one pollster, with the first round inside the margin. In that case the 14.05pp spread stops being a floor and becomes a ceiling, and the compression continues toward single digits. The most direct trigger is AtlasIntel on August 31, with n=5,000, the largest sample in the block: a first-round reading below three points from it would reprice the contract in the same session.
The TSE begins ruling on August 31 on the candidacy registrations of Lula, Flávio Bolsonaro, Romeu Zema and Renan Santos, and Renan Santos’s has already come off the docket through a ruling dated August 29 that pointed to signs of illegality. A decision changing the status of any of the four candidacies moves the outright-winner contract and the placement contracts at once, and the current price does not embed that outcome: Renan Santos’s book, the second largest in the presidential contract with USD 11,880,739 accumulated, pays 2.25%.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| Lula, presidential | 57.50% | -5.00pp | USD 9,353,090 | Sat still at 62.50% through the first three sessions and gave ground in the last two, 4.00pp on Thursday and 1.00pp on Friday. It is his largest weekly drop across the 17 Mon-Fri weeks of the series, which begins on April 20, and the only time he gave up 5.00pp or more in a five-session window. |
| Flávio Bolsonaro, presidential | 37.60% | +4.70pp | USD 9,148,705 | Rose in four of the five sessions, with the largest jump on Tuesday, from 32.90% to 35.20%. It is his largest weekly rise across the 17 Mon-Fri weeks of the series, and it occurs in the same window in which the leader records his largest weekly drop. Both ends of the spread set a record in the same direction, which explains the compression. |
| Renan Santos, presidential | 3.00% | -0.50pp | USD 11,746,020 | His book is the second largest in the presidential contract and the price never passed 3.50% during the week. Accumulated volume and implied probability point in different directions. |
| Renan Santos, first-round third place | 35.50% | -18.50pp | USD 214,417 | Largest absolute move of any contract on the panel during the week. Entered Monday leading at 54.00% and ended Friday tied with Ronaldo Caiado. |
| Augusto Cury, first-round third place | 25.70% | +25.00pp | USD 40,283 | Went from 0.70% on Monday to 25.70% on Friday. The name entered the outright-winner contract on Aug 27, which makes his series short and the comparison with the other rows asymmetric. |
| Flávio Bolsonaro, first-round second place | 87.50% | +2.00pp | USD 410,628 | The market treats his trip to the runoff as close to settled and trades the rest. The contrast with the outright-winner contract, where he pays 37.60%, is the reading that reaching the runoff and winning it are different bets. |
| Impeachment of a Supreme Court justice before 2027 | 3.40% | 0.00pp | USD 83,913 | Did not move in any of the five sessions, with every raw point between 3.40% and 3.50%. It is the smallest book on the panel and the only contract with no weekly variation. |
| PL, most Senate seats | 77.00% | -0.50pp | USD 262,588 | Stability in a contract with a legislative horizon, with the MDB in second giving up 0.90pp, to 15.20%. The whole book adds up to less than USD 300 thousand, and the size caveat applies to both rows. |
| 2026 CPI in the 5.00%-5.49% band | 47.10% | -3.10pp | USD 13,267 | The weight that left the central band went to the one below, 4.50%-4.99%, which rose from 23.10% to 27.50%. The 6.00%-6.49% band gave up from 2.30% to 1.40%. |
5.Liquidity and market structure
The five largest books in the presidential contract belong to names paying between 0.05% and 2.25%, and together hold 2.45% of implied probability against USD 57,151,597 of accumulated volume, or 40.9% of everything traded. Lula and Flávio Bolsonaro, who together hold 96.95% of the probability in the confirmed reading of Aug 30, come sixth and seventh by volume, with USD 9.50M and USD 9.31M. Accumulated volume records where the money has passed since the contract opened, not where it sits: the four names at the top were tested as presidential alternatives in April and May and have not been priced above 1% since.
Accumulated volume is informative and not directional. Book depth is not cited in this publication, by an editorial decision of 21/May/2026: in a continuously arbitraged market, low liquidity does not imply a wrong price, and exposing the number without that context produces a skewed reading.
6.Calendar of price-relevant prints
| Date | Sample | Why it matters | |
|---|---|---|---|
| Aug 31 | AtlasIntel | n=5,000 | Largest sample in the block and the reading most able to move the contract in a single session. Protocol BR-07972/2026. |
| Aug 31 | Real Time Big Data | n=2,000 | Fieldwork from Aug 26 to 29, closed before the weekend. Protocol BR-03733/2026. |
| Aug 31 | Nexus | n=2,000 | A pollster with a series on the panel since July, which allows internal comparison. Protocol BR-08900/2026. |
| Sep 2 | Quaest | n=2,004 | First national Quaest since Aug 14. Protocol BR-07065/2026. |
| Sep 3 | PoderData | n=3,000 | Sample 25% larger than the same pollster’s Aug 27 reading. Protocol BR-07561/2026. |
| Sep 3 | Datafolha | n=2,002 | First presidential Datafolha after the candidates’ TV Globo interviews. Protocol BR-03669/2026. |
Registration with the TSE is not confirmed release, and pollsters may delay or cancel. The table lists six of the eight national polls scheduled in the window, all with a sample of 2,000 or more and national scope; 65 state-level or undetermined-scope registrations from the same period were left out. No number from these polls enters the panel before it is out.
7.Watch list, week triggers
- AtlasIntel on August 31, with n=5,000 is the largest sample registered in the window and the only one able to reprice the contract in a session on its own. A first-round reading below three points would consolidate the direction the price has been taking for four weeks.
- If two of the eight national polls repeat Gerp’s inverted runoff sign the August 26 reading stops being an isolated point and becomes the second observation of a pattern. Today Gerp is the only one of the four in the window that puts the runner-up ahead.
- The ruling on candidacy registrations at the TSE, starting August 31 reaches four of the panel’s names at once. Renan Santos’s has already come off the docket through a ruling dated August 29, and his outright-winner contract did not react on either of the two following days.
- The tie between Ronaldo Caiado and Augusto Cury in the third-place contract closed Friday at 35.50% each, after the week opened with Renan Santos leading at 54.00%. It is the contract with the largest absolute move on the panel and the smallest volume among those tracked, with USD 752,901 in the whole book.
- The downward shift along the CPI ladder took 3.10pp out of the central band and put 4.40pp into the 4.50% to 4.99% band. If the week’s national polls come with a better approval reading for the government, the inflation contract and the presidential one start pointing the same way for the first time since July.
8.Methodology
AFOS does not weight polls or build a pollster average. It reports the prediction market price, each national poll registered with the TSE alongside its sample, margin and reliability rating, and press coverage, leaving visible where the three disagree. Every published price passes through two independent readings eight minutes apart and only goes out when they agree within 0.20pp; the guard certifies book by book, and a book that fails is not published.
The weekly variations in this issue use each day’s close. This week the criterion does not change the result, because Monday had a single capture in the presidential contract and all three possible criteria return the same 9.70pp. The Δ between this issue’s confirmed reading and the previous one is a different window, timestamp to timestamp, and it is reported separately in the first card. Superlatives are checked against the raw points of the series, not against the close-by-close series, and always with the window stated: the 9.70pp compression is the largest of the 17 Mon-Fri weeks since April 20, 2026. The code is open under Apache 2.0 and lives on GitHub.
9.Additional reading · macro coverage
AFOS is the primary source of this analysis. The references below are complementary reading for macro context and do not support any of the measurements above.
- Valor Econômico · Coverage of the electoral calendar’s effect on domestic assets, with the stock market and the real under pressure during the week.(paywall), valor.globo.com/
- Folha de S.Paulo · Coverage of the presidential candidates’ TV Globo interviews and of the start of the candidacy registration rulings at the TSE.(paywall), www1.folha.uol.com.br/poder/
- Poder360 · Record of the August 29 ruling that pulled Renan Santos’s candidacy registration off the docket, with the grounds cited by the rapporteur., www.poder360.com.br/poder-eleicoes-2026/toffoli-adia-a-analise-de-candidatura-de-renan-santos/
- Reuters · International coverage of the Brazilian electoral cycle and of flows into emerging market assets., www.reuters.com/world/americas/
- Gazeta do Povo · Coverage of the proposals moving through Congress that alter the Supreme Court’s powers., www.gazetadopovo.com.br/republica/propostas-para-frear-stf-se-multiplicam-no-congresso-mas-travam-sob-alcolumbre/
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.