US 2026: prediction market, polls and press in the midterm elections

Summary

From April to November 2026 AFOS recorded, side by side, Polymarket prices on control of the House and the Senate, national House vote polls (the so-called generic ballot) and the coverage of a fixed list of outlets, in a public dashboard, a weekly edition for voters and a weekly technical reading. This report brings these series together and sets them beside the result. It describes what each instrument showed; it does not rank the instruments.

1. The Senate market changed favorite; the House market did not. The Democratic Senate contract closed below 50 on 99 of 199 days since Apr 14, with the lowest close, 41.50, on Jun 29, and closed ahead from Aug 30 onward. The House contract never closed below 85.50 from the start of its series, on Jul 29.
2. The polls had the Democrats ahead the whole time, in a narrow range. The AFOS average, as published each day, stayed between D+5.00 and D+8.11 from Jul 30 to Nov 4. On the eve it was D+7.66; on election day, D+7.73, with polls ranging from D+3 to D+12.
3. Market and polls pointed the same way, and the result went that way. On the eve, the contract paid 94.50 on the Democrats in the House and 76.50 in the Senate, and the poll average was D+7.66. The Democrats won both chambers. One result does not grade a price: 76.50 left about a one-in-four chance for the other side.
4. The Senate seats market got the range right. At the Nov 2 close, the brackets summed to 104.30%, within our 95 to 105 gate, and put 57.50% on 47 Republican seats or fewer. There are 44 projected, and at most 46 if Alaska and Mississippi go to the Republicans.
5. On election night, the Senate price reached 0.99 before the networks projected control. At 6 p.m. in New York, when the first polls closed, the Democratic Senate contract was at 0.76. At 9:54 p.m., when South Carolina was projected, it was at 0.992. The networks projected Senate control at 11:38 p.m.
6. The money that reached the count on each side met opposite fates. Among the 180 highest-weight wallets with a verified result, the 42 that reached the count with exposure in favor of the Democrats all won, totaling US$ 2.92 million; the 39 that reached it in favor of the Republicans all lost, totaling −US$ 3.75 million.

The US has no mandatory poll registration, unlike Brazil, where the TSE registers each one. So the AFOS average states its rule: it counts only the question with the party label. Polls that give the names of the district candidates, or that ask which party the voter prefers to control Congress, are listed in the data file and left out of the average, by a decision recorded on Nov 2.

The two contracts told different stories. The House contract was the most stable of the cycle: it opened the series at 85.50, rose one point at a time to 92.50 on Sep 20 and ended the eve at 94.50. The Senate contract was the one that moved: it started at 54.50, fell to 41.50 on Jun 29, returned to 50 in August and, from Aug 30 on, closed only with the Democrats ahead, up to 76.50 on the eve, the highest close in the series.

Before the AFOS series began, the flow study, which reads the platform's own history since the markets were created on Jul 11, 2025, records the Democratic House contract at 0.68 at the open, with an hourly low of 0.565, and the Senate contract at 0.29, with a low of 0.235.

The largest one-day moves in the Senate

The right-hand column says what was public, not the cause of the move. The order in time and the size are measured; the causal link is not.

The third published contract, on the election being held on schedule, stayed between 95.50 (Aug 18) and 99.70 (Nov 2) and resolved yes. The five published prices totaled US$ 96.22 million traded through resolution, according to the platform's cumulative volume.

The polls gave the Democrats the lead throughout the period, and the average moved little: from D+6.82 on Jul 30 to D+5.00 in early September and back to near D+8 in October. The September move coincides with the rise in the Senate contract (section 2).

Between the eve and election day six polls came in (John Zogby Strategies, Reuters/Ipsos, Focaldata, Big Data Poll, YouGov and Rasmussen Reports) and one left, Hart Research with Public Opinion Strategies, which asked which party the voter prefers to control Congress.

The two instruments measure different things, and AFOS never subtracts one from the other. What can be compared is direction and timing. On direction, the two agreed throughout the second half of the year: polls with the Democrats ahead, control contracts with the Democrats favored in the House since the start of the series and in the Senate since Aug 30.

Prices: daily close, except the eve, which is the certified reading. Average: as published that day. They are different objects; the table shows the path of each, not a distance.

On timing, they parted. The poll average stood practically still in October, near D+7.5, while the Senate contract rose from 62.50 to 76.50. The same national lead can be worth very different things for each chamber, because the 2026 Senate was decided in a few states and the House in hundreds of districts, and the Senate seats market showed this: on the eve, it put 57.50% on 47 Republican seats or fewer.

The first polls closed at 6 p.m. in New York, with the Democratic contract at 0.95 in the House and 0.76 in the Senate. At 8:23 p.m., when North Carolina became the first Senate seat projected as flipped, the Senate was at 0.975. When NBC projected the House, around 10:08 p.m., both were already above 0.99. Between 5:30 p.m. on Nov 3 and 12:30 a.m. on Nov 4, there were 18,689 aggressor (taker) orders, totaling US$ 15.36 million, from 3,364 wallets.

The AFOS capture lock certified the Democrats in the House at 98.20 at 00:35 UTC (7:35 p.m. in New York). In the Senate, the book moved faster than two readings eight minutes apart could confirm, and the lock only certified 98.65 at 01:52 UTC (8:52 p.m. in New York), after four attempts. No uncertified reading was published.

Using Polymarket's public data, AFOS reconstructed who traded the three control markets (House, Senate and the "balance of power" market, which pays on combinations of the two chambers) from their creation, on Jul 11, 2025, to resolution. The rule is the same as in the Brazil studies: flow is not impact, and the study does not identify people or attribute intent, coordination or prior knowledge.

October was the highest-volume phase, with US$ 29.79 million. Election night, from 6 p.m. to midnight on Nov 3, moved US$ 14.71 million in six hours, almost as much as the whole eve and election day up to 6 p.m., and had 1,098 wallets that had never traded in these markets.

Who won and who lost

The study looked closely at 183 wallets, chosen by a fixed rule (those with the largest balance and volume in each phase and over the period), and verified the result of 180. Of these, 88 won, totaling US$ 3.65 million, 90 lost, totaling −US$ 5.09 million, and 2 ended at zero. The position each one held at 6 p.m. on Nov 3 explains almost everything:

Patterns assigned by rule, in the study's order; four other smaller categories total 65 wallets. In two rows the number that won is counted over the wallets with a verified result.

The largest gain, of US$ 807 thousand, went to a wallet that debuted on Sep 11, 2026 and made 15,183 trades in 38 days, with 72% of its volume in passive orders. It bought US$ 13.86 million, mainly shares that paid if the Republicans did not end up with the House or with both chambers, at prices from 0.91 to 0.94. At 6 p.m. on Nov 3, it stood to take US$ 798 thousand if the Democrats won both chambers and to lose from US$ 7.26 million to US$ 8.52 million if the Republicans ended up with the House. It risked a lot to earn little per share, and the night marked US$ 569 thousand of that result.

The largest loss, of −US$ 753 thousand, came from a wallet that debuted on Oct 31, three days before the election, and in four days bought US$ 753 thousand of a single type of share: the one that paid if the Democrats did not win both chambers, at prices between 0.30 and 0.23. It would have taken US$ 1.98 million in any of the other three scenarios. It lost everything it put in.

The press entered the series as a record of what was published each day, not as a forecast, and AFOS does not classify outlets by leaning. Three themes ran through the cycle in the 14 Weekly editions: House redistricting in several states, with rulings from state courts and the Supreme Court; the dispute over certification of the Alaska Senate result, with a Justice Department suit rejected by a federal judge on Nov 1; and, in the last week, early turnout, with more than 50 million votes cast before the day, according to the Guardian.

Two rules stayed fixed: a high-impact fact enters only with two sources from different groups, and every claim in the Weekly carries the outlet and the date.

• Price. Published only after two readings agree; the closing series uses the last point stored each day, and the series stores a point when the price moves. Cumulative volume comes from the platform; the flow study volume counts only aggressor (taker) orders, and the two do not add up.
• Polls. Public index checked against the primary source. Different polls do not form a series: methodology, sample and question vary. The rule for the average changed over the cycle, always with a date and a reason, and figure 2 shows each day's average under that day's rule.
• Result. Network projections, not certification. The national House vote is still being counted, especially in the West.
• Wallets. Polymarket public data. The study universe is 183 of 41,569 wallets; there is no order book history; call times taken from aggregators are marked as approximate.

Agreeing is not confirming, and being right is not either. In Brazil, a month earlier, market and polls pointed the same way in the runoff and the ballot box went the other way. In the US, the two pointed the same way and the result went that way. Each case is a single result. What makes it possible to say how far each instrument strayed, and when, is having the record of all three published before the result, with date and time.

A national margin does not tell the size of each chamber. The poll average stood almost still in October, and the Senate contract rose 14 points in the same month. The seats market showed why: Senate control depended on a few state races, and the same national lead was worth more or less depending on them.

The price is fast, not infallible. On election night, the Senate contract went from 0.76 to 0.99 in less than four hours, before the networks projected control. On the eve, the same contract still left a one-in-four chance for the Republicans. It reacts quickly to what arrives; it does not know in advance.

The side decided the money. Among the highest-weight wallets, all that reached the count in favor of the Democrats won and all that reached it in favor of the Republicans lost. The largest loss came from a wallet that entered three days before the election; the largest gain, from one that risked millions to earn cents per share.

The data are in the AFOS-Analytics1/usa-2026-midterms-divergence dataset on Hugging Face: contract and distribution prices, polls with their primary source, the press archive and the distribution coherence gate.

1. Polymarket, Gamma API, Data API v2 and CLOB.
2. Wikipedia, "2026 United States elections polling", as an index; each pollster's release as the primary source.
3. NBC News, House and Senate results, updated at 1:33 p.m. (New York) on Nov 6, 2026.
4. AFOS Weekly, editions No. 1 to No. 14 (Aug 6 to Nov 5, 2026); AFOS Tradeoff US, No. 1 to No. 15.
5. AFOS, flow study of the congressional control markets (data through Nov 6, 2026), internal use, with independent review.

AFOS Analytics · closing report on the 2026 US midterm elections · Nov 6, 2026. This document is a descriptive analysis of public data. It is not an election poll, it is not investment advice and it does not allege coordination, manipulation or wrongdoing. Corrections: contact@afos-analytics.com.

What happened after the count

This part records the reactions to the result in the three days after the election: from the president, the congressional leadership, foreign governments, the press outside the US, the markets and polling experts. Each statement comes with who said it, where and when, as published by the cited outlet. Almost all came through outlets, not through the original channel; when a statement was translated, this is stated.

In the early hours of Nov 4, by phone to Fox News, President Donald Trump accepted the result: "Yeah, sure, I'll accept. Why wouldn't I?" He said he was "a little bit surprised" and attributed part of the defeat to the war with Iran and the price of gasoline (ABC7, ABC News blog, Nov 4). On Truth Social, he wrote that he was not on the ballot and that, if he had been, "we would have won in a landslide" (Euronews, Nov 4).

The same day, according to AFP, he said he would do it all again in Iran, called "largely true" Vice President JD Vance's reading that the war cost them Congress, "because it drove the gasoline up", and said he will get along with the Democrats (France 24, Nov 4). The White House posted a photo of him in the Oval Office with a caption that the Italian agency ANSA translated as "warning: he is still your president" (ANSA, Nov 5; original text not seen).

On Nov 5, in the Oval Office, Trump said "Our elections are so crooked, you have no idea" (France 24/AFP, Nov 6; Il Sole 24 Ore records "so rigged"). In the sources read, the remark is not tied to the congressional races: France 24 links it to the Los Angeles mayoral result, and Il Sole says he did not specify the election. The same day, on Truth Social, he asked that the votes cast for other candidates in Alaska, among them a minor candidate also named Dan Sullivan, be "immediately given to the REAL Dan Sullivan" (France 24, Nov 6). The Alaska Senate race still had no projection.

The reaction from abroad split into three tones: declared relief in Western Europe and Canada, caution from those who depend on the relationship with Washington, and the reading that nothing changes in Moscow, Beijing and Tehran.

No statement on the result was found from Emmanuel Macron, Narendra Modi, Anthony Albanese, Xi Jinping or Vladimir Putin.

Elon Musk, whose spending in the election outlets estimate at between US$ 442 million and nearly half a billion dollars, wrote on X in the early hours of Nov 4: "Guess my election-interfering space lasers didn't work this time" (Click2Houston, with AFP). Tesla and SpaceX fell more than 3% that day (CNBC). Pope Leo XIV voted by mail, according to the Vatican, and spoke of the duty to vote at the Nov 4 audience, without commenting on the result.

The headlines repeated three words: "tsunami", "rejection" and "setback". AFOS does not classify the outlets; the table records what each one published, on open pages that were read. Some of the major European outlets (Guardian, BBC, Financial Times, Le Monde, Der Spiegel, El País) block automated reading, and their absence here does not mean they did not cover it.

On Wednesday, Nov 4, US stocks rose sharply, and the Dow Jones gained about 1,500 points, while oil fell more than 8%; CNBC linked the drop in oil to the administration's acknowledgment of the weight of the war with Iran in the defeat. On Thursday, Nov 5, there was a reversal led by technology and artificial intelligence companies: the Dow fell 1,833 points and the Nasdaq 5.5% (CBS News, Nov 6).

Prediction markets became news in their own right. CNBC, which discloses a commercial relationship and a minority stake in Kalshi, published on Nov 4 that they "cement their status", with Polymarket saying it had anticipated ten key races a median of 31 minutes before the TV networks; the same article recalled that these markets got the Brazilian runoff wrong, and quoted Michelle Kendler-Kretsch, of the Anti-Corruption Data Collective: "Risks and vulnerabilities didn't change overnight just because the markets were accurate this time around." Kalshi says it traded more than US$ 2.5 billion on the midterms, a figure from the company itself and not comparable with those in this report.

The debate over the polls began the next morning. Fox News summed it up in its Nov 4 headline: this time, the underestimated side was not the Republicans. In the same article, the two partners in the Fox News Poll agreed on direction: Democrat Chris Anderson said that "Polls probably, on average, undercounted Democrats significantly", and Republican Daron Shaw that "If we're talking about the direction of the polls, they were clearly correct" and that "we probably underestimated Democratic vote share by a couple of points" (Nov 4). Newsweek measured, against the AP count as of the afternoon of Nov 4, gaps of up to 9.5 points between the final aggregator average and the result in state races, such as the Texas and Georgia Senate races.

AFOS does not measure poll accuracy by state race, and does not set its national average against a partial count. The election-day average was D+7.73 and the eve average D+7.66; the comparison with the national House vote comes with the official count, in version 1.1 of this report.

This is version 1.0, closed on Nov 6, 2026. State certification, the Alaska and Mississippi Senate races and the national House vote were still open. Version 1.1 will bring the official result, the comparison of the poll average with the national vote and any correction, with the old and the new value.

The projected result

NBC News figures, updated at 1:33 p.m. (New York) on Nov 6, 2026. These are network projections, not certification: the states certify through November and December, and version 1.1 will bring the official result.

Sources: NBC News, House and Senate results (nbcnews.com/politics/2026-elections), Nov 6, 2026, 1:33 p.m. ET; projection times: AFOS flow study, with the first publication found for each call; contract resolution: Polymarket.

Who won

The Democrats won control of the House and the Senate. The networks projected the House around 10:08 p.m. on Nov 3 (NBC News) and the Senate around 11:38 p.m. (ABC News), New York time; the AP called the Senate around 11:40 p.m. and the House between 1:25 a.m. and 1:51 a.m. on Nov 4. The Polymarket contracts resolved on Nov 4, the House at 13:34 UTC.

House of Representatives

Senate

What AFOS recorded on the eve, beside the result