AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
BrazilThe week of 21 to 25 September was a round trip in the presidential contract. At 18:00 UTC on Monday the gap between Flávio Bolsonaro and Lula reached 23.80pp, the largest in the entire paired series recorded since 14 April, 7.50pp above any earlier instant. Over the next four sessions it gave ground and closed Friday at 14.90pp. Within the week, from Monday's close to Friday's, that is 5.80pp less, the seventh largest shift in absolute value of the 24 weeks with four or more sessions. Against the close that issue №18 published, 14.00pp on 18 September, the distance ended only 0.90pp higher.
The pullback is not an effect of the pair's sum. The sum stayed almost still, from 99.70 to 99.90, and the normalized reading gives 2.92pp to each side: Flávio Bolsonaro from 60.20% to 57.40% and Lula from 39.50% to 42.50%. The intra-week low of the gap, 10.70pp, came at 15:30 UTC on 24 September, the day six national polls were released, and the range between high and low was 13.10pp in four days.
The polls moved much less than the contract. The ten national polls in the window repeat the previous week's split in the second round, with six putting Flávio Bolsonaro ahead, three putting Lula ahead and one tie, but the interval between the extremes fell from 14.3 to 7.7 points. Six pollsters published in both windows, and none moved more than 1.3 point against itself: three moved up to one point in Lula's direction, two stayed still and one moved 1.3 point in Flávio Bolsonaro's direction. These are different units, a point of voting intention against a percentage point of probability, and the comparison holds as an order of magnitude. This is the last full week before the first round, on 4 October, and the calendar registers seventeen national polls through 2 October.
1.Executive Summary
The week's theme is the distance between how much the contract moved and how much the polls moved. The gap travelled 13.10pp between high and low in four days and returned close to where the previous week had closed. The six pollsters that measured themselves in both windows, with about a week between field dates, changed by at most 1.3 point each, and on average by less than half a point. The split between pollsters keeps the same direction as the previous week, a simple majority for Flávio Bolsonaro in the second round and a majority for Lula in the first, but it narrowed. What the week adds is the scale of the price move without a corresponding move in the pollsters that repeated the question.
2.Why AFOS does not smooth
The ten national polls released between 21 and 25 September asked the same second-round pair, with samples of 2,000 to 5,015 interviews and field dates between 15 and 23 September. An average of them returns a little more than one point to one side, a number that reads as a tie and that erases two things: that the majority points to one winner and three pollsters point to the other, and that the interval between the extremes is almost double the sum of their margins.
Why it matters: the pollsters at the extremes declare margins of 2.2 and 2.0 points, whose sum is 4.2, and the interval between them is 7.7 points, or 1.8 times larger. The interval fell almost by half in one week and still cannot be explained by sampling error, and the average of 1.3 point describes a tie that six of the ten pollsters did not measure.
Issue №18 assigned about 55% to the base case, in which the national polls of 21 and 24 September would keep the split, with a simple majority giving the second round to the price leader, dispersion between pollsters would stay above 10 points and the gap would settle between 12pp and 20pp. The split held, six to three with one tie. The dispersion did not: it fell to 7.7 points. And the band got four of the five closes right, between 14.00pp and 17.20pp, with Monday's at 20.70pp, above the ceiling; in the recorded readings the gap left the band on both sides, at 23.80pp on Monday and at 10.70pp on Thursday. About 30% went to the contrarian case, in which the six national polls of 24 September would repeat Lula's side in the second round as a block. It did not happen: of the six, only Datafolha put Lula ahead, and the day's three largest samples gave Flávio Bolsonaro by 3 points, Flávio Bolsonaro by 1 point and a tie. The gap closed 5.80pp lower within the week without that trigger. About 15% went to the tail, the Supreme Court impeachment contract closing a day above 8%. It did not happen, not even as a touch: the week's high was 4.40%, and the contract closed Friday at 3.40%. Of that issue's calendar, nine of the ten pollsters listed published in the window, and the national label of Real Time, which the issue gave as under review, was confirmed at the release of 24 September.
3.Weighted scenarios for the week
The week of 28 September to 2 October is the last before the first round, on 4 October, and concentrates seventeen registered national polls: Quaest and NEXUS on Monday, seven registrations on Tuesday, two of them from Gerp with field dates after the release date, three on Wednesday, Datafolha and Real Time on Thursday and three on Friday. On Thursday, 1 October, TV Globo holds the last debate before the vote.
The last wave repeats the shape of the two previous weeks: a simple majority of the national polls giving the second round to Flávio Bolsonaro, a majority giving the first to Lula, and pollsters measuring against themselves moving by up to two points. In that case the gap stays within the band travelled this week, between 10pp and 24pp, and the book goes on without a direction sustained for more than two sessions. The intraday range stays high, because seventeen prints in five days give the price a new reason to reprice almost every session.
The week's three heaviest pollsters, AtlasIntel on 29 September, with 5,000 interviews, Quaest on the 28th and Datafolha on 1 October, put Lula ahead in the second round at the same time. AtlasIntel already measured Lula 0.3 point ahead in this window, Datafolha measured Lula 2.0 points ahead in two rounds of polling running and Quaest stood 1 point on the other side. If the three converge, the contract has to choose between the Flávio Bolsonaro advantage it has sustained since the September inversion and the three heaviest readings, and this week's low, 10.70pp, stops being the low.
The last wave comes to show one of the two with a majority of valid votes in the first round outside the margin, at more than one pollster. The winner contract today prices, in practice, the second round; an election settled on 4 October would change the question it answers and the date on which it settles. None of the ten national polls in this window measured that: in the first round the largest distance between the two was 7 points, with both below 46%.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| Presidential gap, Flávio minus Lula | +14.90pp | -5.80pp | USD 155.19M | Path: +20.70, +17.20, +16.70, +14.00, +14.90. High of 23.80pp at 18:00 UTC on Monday, the largest in the paired series, and low of 10.70pp at 15:30 UTC on Thursday. |
| Flávio Bolsonaro, presidential | 57.40% | -2.80pp | USD 11.86M | High of 61.70% at 17:30 UTC on Monday, the largest in his series, and low of 54.30% at 03:30 UTC on Friday. Took in USD 479 thousand over the window. |
| Lula, presidential | 42.50% | +3.00pp | USD 12.26M | Low of 37.50% at 18:00 UTC on Monday and high of 44.50% at 15:30 UTC on Thursday. Took in USD 555 thousand, the book's largest single inflow. |
| Flávio Bolsonaro, 2nd place in the first round | 74.50% | +5.00pp | USD 1.16M | It rises while he falls in the winner contract: the book now prices Flávio Bolsonaro finishing second in the first round more heavily. Low of 67.50% on Monday and high of 77.50% on Friday. |
| Lula, 2nd place in the first round | 25.40% | -3.20pp | USD 0.99M | The mirror of the two previous weeks did not repeat: +5.00pp on one side against -3.20pp on the other, with the book's sum rising from 105.30 to 107.10. |
| Renan Santos, 3rd place in the first round | 45.00% | -0.50pp | USD 0.41M | An almost unchanged close over a range of 11.50pp, from 52.50% at Monday's open to 41.00% on Tuesday. Took in USD 51 thousand. |
| Augusto Cury, 3rd place in the first round | 42.00% | +1.70pp | USD 0.28M | Low of 32.40% at 06:00 UTC on Tuesday and high of 43.20% at 17:30 UTC on Friday. Ended the week 3.00pp behind Renan Santos. |
| Ronaldo Caiado, 3rd place in the first round | 14.50% | -1.00pp | USD 0.17M | The whole book holds USD 1.31M accumulated and took in USD 105 thousand, and the sum of the outcomes closed at 111.00, so price here calls for a normalized reading. |
| Supreme Court impeachment before 2027 | 3.40% | -0.90pp | USD 0.09M | Range of 1.00pp, between 3.40% and 4.40%, on USD 76 of new money over the whole week. |
5.Liquidity and market structure
The Renan Santos contract overtook Tarcisio de Freitas as the book's largest accumulated volume between the closes of 18 and 25 September, from USD 13.69M to USD 14.45M, and closed the week paying 0.60%, after giving up 0.40pp. The book took in USD 2.18M of new money over the window. Lula and Flávio Bolsonaro took USD 555 thousand and USD 479 thousand; right behind come Renan Santos, with USD 452 thousand, Geraldo Alckmin, with USD 365 thousand at 0.20%, and Jair Bolsonaro, with USD 113 thousand at 0.10%. The three add up to USD 930 thousand, more than two fifths of the week's new money, in contracts paying 0.60% or less. Accumulated volume is trading history since each contract opened and is not current conviction, and anyone who reads the volume ranking as a probability ranking is reading the book backwards.
Volume is accumulated since each contract opened. The week's new money is the difference between the volumes recorded at the closes of 21 and 25 September, and the prices in the table are those of 25 September. AFOS does not cite book depth in editorial text, because low liquidity in a continuously arbitraged market does not mean the price is wrong.
6.Calendar of price-relevant prints
| Date | Sample | Why it matters | |
|---|---|---|---|
| 28/Sep | Quaest and NEXUS | n=2,004 and n=2,000 | Both opened the two previous windows. Quaest's field dates run from 24 to 27 September and straddle the provisional measure banning betting sites, published on the 25th. |
| 29/Sep | AtlasIntel, Palver and JOTA | n=5,000, n=5,000 and n=6,000 | The week's three largest samples. In this window AtlasIntel measured the narrowest distance in the second-round pair, 0.3 point, with the smallest declared margin, 1.0; Palver measured 4 and 3 points for the other side in two rounds of polling. |
| 29/Sep | Vox Brasil and Real Time Big Data | n=2,100 and n=1,600 | Real Time returns with the national label coming from the sampling plan, which the house repeats in its state polls; the 24 September registration was confirmed as national at release. |
| 30/Sep | Futura/100% Cidades, Ideia and Indexa | n=2,000 each | Futura measured 5.7 points for Flávio Bolsonaro in this window, outside its own margin, and is the upper end of the interval. |
| 01/Oct | Datafolha | n=2,506 | Field dates of 28 September to 1 October, ending on the day of the TV Globo debate. The only one of the six national polls of 24 September to put Lula ahead in the second round. |
| 02/Oct | Vox Brasil, Datavero and DataTrends | n=2,100, n=1,500 and n=1,200 | The last national polls registered before the vote on 4 October. Datavero has a single registration in the database and its national label comes from the sampling plan. |
Registration with the electoral court is not confirmed release. Seventeen national polls with samples of 1,000 or more are scheduled for release from 28 September to 2 October, and the table shows fifteen: Real Time's second line, with 2,000 interviews on 1 October, stays in this footer, and two Gerp lines with release registered for 29 September are left out, because their field dates only begin on 30 September and 1 October, after the date the registration declares.
7.Watch list, week triggers
- The gap at 14.75pp and third place changing hands over the weekend The confirmed reading of 27 September at 15:47 UTC puts Flávio Bolsonaro at 57.25% and Lula at 42.50%. In the third-place book, Augusto Cury overtook Renan Santos, 44.40% to 41.50%, outside this issue's window.
- The pollsters measuring against themselves Quaest, NEXUS, AtlasIntel and Datafolha, four of the six pollsters that published in the last two windows, return this week. Between those two windows none of the six moved more than 1.3 point in the second round; a larger move in the last wave would be the first shift in the electorate measured by the same yardstick.
- The betting measure inside the field dates The provisional measure banning betting sites came out on 25 September. The Quaest poll of 28 September is the first national poll with field dates on both sides of that date, and those from the 29th onward have field dates entirely after it.
- The TV Globo debate on 1 October The broadcaster invited Lula, Flávio Bolsonaro, Augusto Cury and Ronaldo Caiado, and Renan Santos was left out under the legal criterion of congressional representation. Lula and Flávio Bolsonaro had not confirmed attendance as of 27 September. Datafolha closes its field dates the same day.
- Priced inflation rose half a point of band The 5.00% to 5.49% band moved to 45.60% and the 6.00% to 6.49% band to 6.40%, in a book of USD 131 thousand accumulated that took in USD 11 thousand. It is a thin book, and the upward shift of mass happened without relevant flow.
8.Methodology
AFOS does not produce a weighted average and does not arbitrate between sources. It publishes the three signals side by side, contract price, registered poll and news flow, and reports the distance between them when it exists. A market price goes live only after two independent readings eight minutes apart that agree within 0.20pp, contract by contract, and the reference series are measured in the daily backup of the database and not in the public interface, which caps the window at ninety days. The gap in this issue is measured at paired instants, both sides with the same timestamp, and every claim of an extreme declares the window in which it was checked.
The polls cited are those registered with the Superior Electoral Court, with protocol, sample, margin and field dates declared, and national scope is a condition of entry into the comparative table. When the scope label depends on a text field that the house repeats without distinguishing, the issue says so in the line instead of publishing the label as fact. The code and the method are open under the Apache 2.0 licence. Every factual claim in this issue links to its primary source.
9.Additional reading · macro coverage
Complementary macro context. AFOS is a primary source on divergence between market and polls; the links below cover the economic and institutional backdrop of the window.
- 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 non-resident investor., www.reuters.com/world/americas/
- Poder360 · Registry of electoral polls with protocol and history by pollster., www.poder360.com.br/pesquisas/
- Gazeta do Povo · Coverage of poll releases with verification of the electoral court registration., www.gazetadopovo.com.br/eleicoes/
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.