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 №3·Week of 2026-08-13 to 2026-08-20·published thursday 07:00 BRT

What the prediction market did

The eight midterm contracts hold USD 17.84M, up from USD 17.48M last Thursday. Last week all movement sat in the Senate; this week both chambers moved, against each other. Both books add to exactly 100.00%, so the raw price is the normalized one.

Senate · Democrats

50.50%

↓ 2.00pp on the week

On USD 2.33M. It held 52.50% from August 12 to 16, then stepped down on August 17, 08:01 UTC.

Senate · Republicans

49.50%

↑ 2.00pp on the week

On USD 1.64M. Last Thursday it printed the lowest value of the queryable window; it has recovered that decline exactly.

House · Democrats

88.50%

↑ 1.00pp on the week

On USD 5.36M, the deepest book here. It has printed 88.50% four times, on August 9, August 17 and today, never above; between touches it falls back to 87.50%.

House · Republicans

11.50%

↓ 1.00pp on the week

On USD 4.18M. The stored series holds only 13.50% and 12.50%; today's certified capture reads 11.50%, below anything we hold.

The fifth published price, on whether the election happens on schedule, sits at 95.75%, flat on the week, on USD 0.36M, still the smallest book of the set. The House series opens on July 29, 2026, so every superlative here means since that date.

What the polls did

The disagreement comes before the average, because it is the finding. Twelve institutes measured the same national race in the same window and came back 9 points apart. And the polls have a second thing to say by saying nothing: no institute has published new national fieldwork in sixteen days. The most recent field date in the file is still August 4. The average reads D+5.93, from 14 polls.

Widest

Big Data Poll (R) · D+11

Narrowest

McLaughlin & Associates (R) · D+2

Spread between institutes

9 points

Big Data Poll surveyed 2,617 likely voters through July 29; McLaughlin 1,000 through July 21. Both are Republican-affiliated and they sit at opposite ends, the cleanest evidence that the gap is method, not lean. Among the three most recent, fielded August 2 to 4, the spread narrows to D+4 to D+6. The average fell from D+6.34 with no new poll: the source returned the same 363 rows, the window rolled a day, and Beacon Research/Shaw, Emerson College and Focaldata fell off the far edge, taking the set from 18 polls to 14. Declared rather than counted: Emerson with The Hill fielded August 16 and 17 on 1,000 likely voters, publishing D 50.7% x R 43.0%, absent from the index we read. We do not hand-add rows: a hand-added number cannot be reproduced by the next reading.

What the press reported

What the fixed-list outlets covered, August 14 to August 20

The week's events were primaries, held Tuesday, August 18, in Florida and Alaska, with results through August 19. The Guardian and The Hill reported that Angie Nixon, a democratic socialist, won Florida's Democratic Senate primary in an upset and will face the Republican incumbent, Senator Ashley Moody. The New York Times reported that David Jolly, a former Republican, took the Democratic nomination for governor.

In Alaska, CBS News reported on August 19 that Senator Dan Sullivan and Mary Peltola advanced, a matchup state polling had called a nail-biter.

Senate races dominated coverage beyond those states. The Washington Post reported on August 14 that Michigan's race was close in the first poll after Abdul El-Sayed's primary win, and Politico asked on August 16 whether he could cost Democrats the state. One item was about the instrument, not the contest: The Hill reported on August 19 that prediction markets were left out of a White House meeting with tech executives.

Our measurement: Both are right and not the same measurement. Those eight points are one poll, Emerson with The Hill, fielded four days ago. Our D+5.93 averages fourteen polls from twelve institutes over thirty days, the newest fielded August 4. That single reading, at D+7.7, sits inside our published spread of D+2 to D+11 and above our average.

The crossing of the week

The chamber with a national ruler barely moved. The other moved twice as far.

The generic ballot is a House instrument: it asks which party a voter would back for Congress in their own district. There is no national equivalent for the Senate, because only a third of those seats are contested and each is a state race. So the chambers are not equally instrumented. This week the one with a national ruler held the three-point band it has kept for 23 days; the one without moved twice as far, and the week's reporting was almost all about named Senate races.

A race that already resolved, and what each instrument said before

Wisconsin's Democratic primary for governor settled on August 11. Clifford Young, President of Ipsos Polling and Societal Trends, put the mechanism plainly to NewsNation on August 15: *"Polling data feeds prediction markets. When the polls are way off, the markets are too."* That race puts it one notch narrower. The market did not inherit the poll's error. It inherited the poll's average and discarded its uncertainty. Marquette Law School had Francesca Hong at 38% and David Crowley at 7% on 407 registered voters, ±6.6pp, with 34% undecided: a poll saying plainly that the race was not decided. The price the day before said 95.8% for Hong. Crowley won by about 3,200 votes. Marquette got the leader wrong and declared its limits; the market got the leader wrong and declared 96% certainty. Three caveats, and note that this brief prints binary prices too. The field changed: Mandela Barnes held 16% at Marquette and is absent from the result, so part of Crowley's rise is redistribution, not error. The institutes disagreed publicly beforehand, State Navigate finding Hong 44 x Crowley 22 on 1,473 respondents. And that 95.8% is not our capture: it is attributed to the outlet that reported it, linked below.

How to read this number

A price is a point. The poll it read was a distribution.

A fifteen-point lead with 6% undecided and one with 34% undecided produce the same average and mean different things. That is why this brief prints the 9pp spread beside the average, and why we write that the market is paying 88.50% rather than that the probability is 88.50%. The first observes a price; the second is a claim about the world.

Sources

Prediction market

Polling, primary sources

Press, outlets cited in sections 4 and 5

Prices cited in section 5, which are not our own capture

  • Newsweek · Francesca Hong's Chances of Beating David Crowley in Wisconsin: Final Polls, the source of the 95.8% quoted for the day before the vote, www.newsweek.com/

Method

AFOS reads three signals and does not merge them: prediction market prices, published polls, and press coverage. The house average of the generic ballot is a simple mean, unweighted, and it is declared as such on the dashboard. Market prices come from a certified capture, two readings eight minutes apart that must agree inside 0.20pp before anything is published. Band markets are published only when the bands sum between 95% and 105%; the popular-vote-margin market is currently near 151% and is collected daily for the series without being published as a reading. This brief never subtracts a market probability from a polling lead, because they are different quantities. Open source, Apache 2.0.

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. 3 · The two chambers moved against each other, and the only national ruler has not moved in sixteen days | AFOS Analytics