AFOS Weekly · USA-2026 midterms Political Risk Weekly

AFOS Analytics

Real-time political pricing
Prediction market × polls × press
no smoothed averages

Harvard Dataverse · DOI 10.7910/DVN/XRUT8U
United States·Issue №11·Week of 2026-10-08 to 2026-10-15·published thursday 07:00 BRT

What the prediction market did

The five published prices hold USD 24.87M and took USD 2.88M of new money on the week, four fifths of it in the two House books, where the price never moved. The Democratic House book crossed ten million dollars of cumulative volume after Thursday’s close, on the live reading.

House · Democrats

90.50%

→ 0.00pp on the week

On USD 9.99M, with USD 1.38M of new money, the most of any book. All eight daily closes from October 8 to October 15 printed 90.50%, across 44 readings, with a weekly range of 1.00pp. The money arrived and the price stayed.

House · Republicans

9.50%

→ 0.00pp on the week

On USD 7.62M, with USD 0.95M of new money. The mirror, also unmoved for eight closes. The pair sums to 100.00% at both ends, so no spread hides a move: flat raw is flat normalised.

Senate · Democrats

63.50%

↑ 1.00pp on the week

On USD 3.79M, with USD 0.22M of new money. It opened at 62.50% and added a point on October 11. ⚠️ But the pair opened summing to 98.00% and ended at 100.00%, so normalised this side lost 0.28pp: the raw number rose and the implied probability fell.

Senate · Republicans

36.50%

↑ 1.00pp on the week

On USD 2.80M, the thinnest control book, with USD 0.28M of new money. It also rose 1.00pp raw and gained 0.28pp normalised.

The fifth price, on the election happening on schedule, closed at 99.10%, up 0.70pp on the week, on USD 0.67M. That matches the highest point of the series, first reached late on October 13; the series begins on July 29.

What the polls did

The spread comes before the average, because the spread is the finding. The average is D+7.58 over 41 rounds from 29 polling firms, and the rounds inside it disagree by ten points.

Widest

Quinnipiac University and Marist · D+12

Narrowest

Harvard/Harris Poll/HarrisX · D+2

Spread between institutes

10 points

The ten points are not dates: Quinnipiac closed fieldwork on September 27 and Harvard/Harris on September 28, one day apart and ten points apart. Sorted by who was screened rather than by who asked, the 26 likely-voter rounds average D+7.23 and the 13 registered-voter rounds D+8.15. That compares two groups of polls, not the same question twice.

What the press reported

What the fixed-list outlets covered, October 9 to October 15

Money was the week's story, and the outlets did not agree how much of it there was. On October 12 the New York Times and the Hill both put Musk's spending at $276M; the same day Reuters reported $366 million, a total the Guardian carried on October 13. We do not measure campaign finance, so we record both and adjudicate neither. On October 14 the New York Times put ad spending past $10 billion, and on October 15 NewsNation reported AdImpact calling the cycle the most expensive ever.

Forecasters moved one way all week: on October 9 the Washington Post reported Cook shifting 18 more House races toward the left, on October 11 NewsNation quoted a forecaster saying the winds were against Republicans, and on October 15 the Hill reported Brit Hume calling a Republican "washout" possible.

And the instrument this newsletter reads became a subject itself: on October 14 Reuters published "Inside the booming market for US midterm election bets", and on October 10 the Hill reported Kalshi investigating suspicious bets tied to a White House appointment.

Our measurement: A broadcaster changing its own call is not the same object as a contract price, and we call it no contradiction. Two facts sit beside it. The poll Fox News published that week gave Democrats 54 to 45 among 1,202 registered voters and 52 to 46 among those certain to vote. And the House price closed at 90.50% on every one of the eight days around that call.

The crossing of the week

The same pollster, the same week, two answers three points apart

Two of the three rounds that entered the average this week carry a second number the aggregating index did not tabulate. Beacon Research and Shaw & Company, fielded October 9 to 12 for Fox News, published D+9 among registered voters and D+6 among those certain to vote. RMG Research, fielded October 13 to 14, was tabulated at D+8 among registered voters while its own headline reported D+11 among likely voters. Three points each, opposite ways. We serve the tabulated cut and changed nothing: taking the other would mean ingesting a row the index does not carry, which changes where the number came from. Measured, following the likely-voter cut on both moves our average by 0.01pp: they very nearly cancel.

Two prices rose and one of them fell

Senate Democrats and Senate Republicans both gained 1.00pp raw this week. Both sides of a two-outcome market rising together is not possible as a statement about the race, and the pair explains it: it opened the week summing to 98.00% and closed at 100.00%. Normalised, the Democratic side gave back 0.28pp and the Republican side took 0.28pp. What moved was the gap between the two quotes, not the probability.

How to read this number

The likely-voter screen, and why it can move a poll either way

Every pollster this close to an election has to decide who will show up. A registered-voter sample asks everyone on the rolls. A likely-voter sample keeps only those the firm judges will actually vote, using questions about past turnout, intention and interest. That screen is the firm's own model, not a measurement, and the two electorates are not the same people: one that favours high-interest voters helps whichever side is angrier this year. Across our window the lean is mild, likely-voter rounds running 0.92 of a point below registered-voter rounds. The reading for a voter: behind one number there is often a second, and the gap is a modelling choice the headline does not show.

Sources

Prediction market, October 15 closes

Polling, primary sources

Press cited in sections 4 and 5

Method

Market prices in this issue are October 15 closes, the last point recorded that day in our daily database backup, compared with the October 8 closes that Issue No. 10 published. Both ends of each binary pair are printed raw and normalised, because a pair that does not sum to 100 moves one side without the other side trading. The polling average is a simple, unweighted mean of one round per polling firm per wave over a rolling thirty-day window, with the hierarchy likely voters over registered voters over adults when a firm publishes more than one row in the aggregating index. When a firm publishes a second screen that the index does not tabulate, we keep the tabulated row and record the other one in our source-verification file rather than ingesting it, because ingesting it would change the provenance of the served number. We never subtract a market price from a polling margin: one is a probability of controlling a chamber and the other is a margin in vote share, and in 2012 Democrats won more House votes and fewer seats.

AFOS
AFOS Analytics
Global Political Risk Intelligence · GLOBAL BY DESIGN
www.afos-analytics.com · contact@afos-analytics.com · github.com/AFOS-Analytics

Get it by email

AFOS Weekly follows the US midterms every Thursday, written for voters: where the prediction market, the polls and the press disagree, and what each one measures.

No spam. Unsubscribe anytime.

After signing up you choose whether to receive it in English, Portuguese or Spanish.

AFOS Weekly · Issue No. 11 · Two pollsters published both numbers, and the voter screen moved them opposite ways | AFOS Analytics