2026 House Popular Vote Odds: How Big Is The Democratic Margin?
Updated August 19, 2026 · 11 min read · by Jake Hari
Everyone asks who wins the House. Kalshi runs a harder market underneath that one: not which party wins the national House popular vote in 2026, but by how much. It lists ten mutually exclusive bands, from a Republican win at one end to a Democratic margin of sixteen points or more at the other, and traders have to spread their money across the whole range. We put the same ladder to a panel of AI models, price-blind, and the models agree with the market on the easy part and part company on the hard part.
The Quick Answer
The market and the panel agree that Democrats win the national House popular vote, and they even agree on the single most likely result: a margin of six to eight points. Where they split is the shape of everything around that peak. The market gives roughly a 60% chance the Democratic margin lands at eight points or larger; the panel, reasoning only from the national environment, gives about 33%. The panel is actually more sure Democrats win the popular vote at all, 96% against the market's 90, but it prices a smaller, tighter win rather than a runaway wave. The full ladder, every model's number, and the exact triggers that would move it are below.
Free: The Weekly PM Market Brief — the 8-model panel's graded record, the week's biggest market-vs-model gaps, and what's spiking next. One email, Sundays.
New to this kind of market? How election prediction markets work explains what a price actually means before you read one.
The Market
| Venue | Kalshi, a CFTC-regulated exchange (18+; availability varies by state, as of August 2026) |
| The Event | "2026 Midterms: House popular vote margin of victory?" — ten mutually exclusive bands of the Democratic margin, the generic-ballot question priced as a distribution |
| How It Settles | On the actual national House vote from the November 3, 2026 election: the share of all votes cast for Democratic House candidates minus the share cast for Republican House candidates, as one nationwide number, with no rounding |
| Settlement Source | Kalshi's rules define that calculation but name no single data provider in the available contract terms; the number comes from official general-election House vote totals. Kalshi lists an expected expiration of November 3, 2027, a buffer for final certified tallies |
| Prices As Of | August 19, 2026, last-traded levels; the full ladder is in the board below |
| Note On Quotes | Long-dated election contracts trade thinly and often show no live bid or ask between trades. Prices here are the most recent traded level, read at build time. Last-traded prices across a ten-way ladder include the exchange's spread, so the market column sums above 100. |
This is a neutral, data-driven analysis. It presents both parties' paths identically, advocates for neither, and every electoral fact in it was machine-fetched from the named sources at generation time.
Hottest Prediction Markets Right Now
- 2028 Democratic presidential nominee · $179M traded
- FedEx St. Jude Championship Winner · $66M traded
- 2027 Pro Football Champion · $61M traded
- 2028 U.S. Presidential Election winner? · $59M traded
- 2028 Republican presidential nominee · $57M traded
- Pro Baseball Champion · $53M traded
Every market above links to our full AI model verdict; browse them all on the OddsShopper prediction markets hub, and see how every settled call actually scored on the full graded scoreboard.
The Board
Each band is its own contract that pays $1 if the final margin lands inside it. The AI blend is the equal-weight mean of five models, each of which priced the entire ladder in one pass without seeing a price.
| Band (Democratic Margin) | Market | AI blend | Gap |
|---|---|---|---|
| Republicans Win (Margin Below 0) | 10¢ | 4% | +6 |
| Democrats +0 To 2 | 1¢ | 7% | -6 |
| Democrats +2 To 4 | 4¢ | 13% | -9 |
| Democrats +4 To 6 | 8¢ | 20% | -12 |
| Democrats +6 To 8 | 24¢ | 22% | +2 |
| Democrats +8 To 10 | 27¢ | 17% | +10 |
| Democrats +10 To 12 | 21¢ | 10% | +11 |
| Democrats +12 To 14 | 8¢ | 4% | +4 |
| Democrats +14 To 16 | 2¢ | 2% | even |
| Democrats +16 Or More | 2¢ | 1% | +1 |
Read back out of the same ladder:
| Question | Market | Panel blend |
|---|---|---|
| Democrats Win The Popular Vote At All | ~90¢ | 96% |
| The Margin Is Democrats +8 Or Larger | ~60% | 33% |
These are not separate estimates. Each model built one distribution and every summary number is read back out of it, which is why nothing here contradicts itself. Every cell in this piece is rounded to a whole point while the blends are computed on the unrounded numbers, so any single column (a model's or the blend's) can total 99 to 103 rather than exactly 100, and a read-back summary can sit a point off the bands it sums.
More live boards from the same panel: House control, which party wins the chamber (Democrats 86¢) · the full four-way balance of power (the split at 39¢) · Senate control (Republicans 53¢). Prices fetched August 19, 2026.
Every Seat's Number
| Band | Claude Fable | Claude Opus | Claude Sonnet | GLM | DeepSeek | Blend |
|---|---|---|---|---|---|---|
| Republicans Win | 6 | 5 | 3 | 6 | 2 | 4% |
| D +0 To 2 | 9 | 9 | 6 | 7 | 5 | 7% |
| D +2 To 4 | 15 | 15 | 13 | 12 | 10 | 13% |
| D +4 To 6 | 20 | 20 | 21 | 18 | 20 | 20% |
| D +6 To 8 | 20 | 20 | 24 | 21 | 27 | 22% |
| D +8 To 10 | 15 | 16 | 18 | 16 | 20 | 17% |
| D +10 To 12 | 9 | 10 | 9 | 10 | 10 | 10% |
| D +12 To 14 | 4 | 5 | 4 | 5 | 4 | 4% |
| D +14 To 16 | 2 | 2 | 2 | 3 | 2 | 2% |
| D +16 Or More | 1 | 1 | 1 | 2 | 1 | 1% |
| Democrats Win Popular Vote (Derived) | 94 | 95 | 97 | 94 | 98 | 96% |
Every seat on this panel is graded against real market settlements — records to date: Claude Fable 84% on 508 graded calls · Claude Opus 85% on 578 graded calls · Claude Sonnet 84% on 556 graded calls · GLM 81% on 2,367 graded calls · DeepSeek 80% on 2,392 graded calls. Recomputed daily; the full scoreboard is public.
Model estimates generated August 19, 2026, price-blind, from the data card described below, after a revision round in which each model read the others' anonymized reasoning. These are model estimates, not predictions of fact and not financial or trading advice. Models are frequently wrong; the market price reflects real traders' money.
Every number above is stored and graded against the actual result when this market settles. Our running record on every event we have priced sits on the full graded scoreboard. Five of the panel's eight seats returned a distribution within the run window; the blend is the equal-weight mean of those five.
What The Panel Was Given
Three layers, every one machine-fetched at run time. No market prices appeared anywhere in the material the models saw.
- The national frontier. The generic congressional ballot averaged across six public aggregators, and presidential approval averaged across ten, each read from the current published tables on August 17, 2026. The generic ballot sits at Democrats +6.6, and the striking part is the agreement: the six aggregators land between +6.1 and +6.8. Presidential approval averages 37.6% approve against 59.3% disapprove, a net of nearly 22 points underwater.
- The popular-vote ledger. The actual national House popular-vote margin from the five most recent midterms, read live from each cycle's official results: 2006 was Democrats +8.0, 2010 Republicans +6.8, 2014 Republicans +5.7, 2018 Democrats +8.6, 2022 Republicans +2.7. That is the range the panel's recent-cycle ledger covers, from a red wave near six points to a blue wave near nine. Deeper postwar history has run wider (1974 reached Democrats +16.8), but nothing in the last two decades has come close.
- The measured error band. In presidential years, polls overstated Democrats by 3.7 points in 2016 and 4.7 in 2020. In the two Trump-era midterms they were nearly unbiased: 0.3 the other way in 2018, 0.9 in 2022. The models were told to apply the midterm figures, not a presidential-year correction, and not to assume zero error either.
The panel was also handed the postwar base rate, that the president's party has lost House seats in 18 of the last 20 midterms, and the two exceptions with them: 1998, the impeachment-backlash year, and 2002, fought after September 11 with presidential approval above 60. Each exception needed approval far above water or a rally event. With approval 22 points underwater and no comparable shock on the board, both conditions are absent, which is the single fact every model leaned on hardest.
Where The Market And The Panel Split
Start with what they share, because it is most of the ladder. Both put the single most likely result at Democrats +6 to +8, within two points of each other (market 24¢, panel 22%). Both treat a Republican win of the popular vote as a real but small tail. And both agree the far upper bands, a margin above 14 points, are long shots: nothing in the panel's five recent cycles exceeded 2018's +8.6, and while the deeper postwar record does hold blowouts on that scale (1974's post-Watergate landslide at Democrats +16.8), the current environment shows no shock of that size.
The disagreement is about the weight on either side of the shared peak, and it resolves to one number. The market gives a margin of eight points or larger roughly a 60% chance. The panel gives it 33%. Every model priced the +8-to-+10, +10-to-+12 and +12-to-+14 bands below the market, and priced the +2-to-+4 and +4-to-+6 bands above it. The +4-to-+6 band is the clearest single gap on the board: the market pays 8¢, the panel says 20%.
The argument behind that is a translation, not a mood. The generic ballot is a survey of stated preference. The settled market pays on cast votes. The panel's models refused to assume the two are the same and refused to assume the Democratic lead balloons between August and November. They took the +6.6 preference number, shaded it down by roughly half a point for uncontested seats and turnout composition, and centered the actual margin near +6. The market, by contrast, is paying up for the lead to grow into a true wave, an outcome that has happened (2006, 2018 both landed near +8) but that the panel treats as one branch rather than the base case.
The one-line version: both sides agree Democrats win the national House vote. The market is paying for a big win. The panel sees a clear but ordinary one, right about where the generic ballot already sits.
There is no edge on the question everyone actually searches, which party wins the House vote, because both sides answer it the same way. The edge, if the panel is right, is entirely in the size.
Dark Horses The Panel Won't Dismiss
The board is not only its peak. Four bands sit under 10% where at least one model saw a real number worth naming.
- A Republican Popular-Vote Win, 4% Blend. Every model kept a live tail here even while calling Democrats near-certain to win. DeepSeek priced it lowest at 2%, Fable and GLM highest at 6%. It is the band that cashes only on a large, late swing plus a repeat of pro-Democratic polling error the midterm archive does not support.
- Democrats +0 To +2, 7% Blend. The narrowest possible win. It is a 2022-sized margin with the sign flipped, since 2022 went Republican but by under three points, a reminder that even accurate polling can compress to a near-tie on election day.
- Democrats +12 To +14, 4% Blend. The lower edge of true wave country. It needs the environment to widen past where any aggregator currently reads, but 2018 shows it is not fantasy.
- Democrats +16 Or More, 1% Blend. The historic-blowout band. One postwar midterm reached it, 1974's post-Watergate landslide at D+16.8, but nothing in the panel's five recent cycles came within seven points of that. GLM alone gave it as much as 2%; the rest priced it at a single point or less.
Where The Panel Changed Its Mind
Each model priced the ladder once, then read the others' reasoning with the names stripped off and was invited to revise. Revision was optional. The interesting part is that it moved in both directions, which is what a real convergence looks like rather than a stampede toward the crowd.
The largest move came from the model that had opened widest. DeepSeek started with a fat Republican-win tail and a flatter center, then read the base-rate argument and sharpened onto the peak, cutting its Republican-win number by three-quarters:
"The distribution peaks at D6_8, reflecting a stable, high-confidence out-party advantage without assuming a repeat of 2018's D+8.6 wave." — DeepSeek
Claude Sonnet had been the most hedged seat, holding wide tails on both ends. It tightened toward the center and explained exactly why it capped the top:
"I keep the upper tail thin beyond D14 since no midterm in the ledger has exceeded 2018's D+8.6 — a D14+ result would be an unprecedented wave requiring conditions not evidenced in the card." — Claude Sonnet
One seat deliberately did not converge. Claude Fable held slightly more weight on the low bands than its peers, and after reading them said why it was keeping it:
"The gap between an August preference survey and November cast votes is empirically wider than their thin [Republican-win] allocations imply." — Claude Fable
What Would Change The Panel's Mind
The panel's read rests on checkable things, which means specific, dated developments would legitimately move it. These are the ones the models named.
- The Generic Ballot Moving Off +6.6. The aggregators update daily. Tightening below Democrats +3 pushes mass down into the narrow-win and Republican-win bands; widening past +9 is what would start validating the market's wave pricing and fatten the +8-to-+12 bands.
- Presidential Approval Recovering, Or A Rally-Type National Event. Approval climbing back toward the mid-40s, or a shock that rallies the country behind the president, is the 1998 and 2002 mechanism. It is the only development on this list that weakens the base rate the whole ladder rests on, and it would fatten the low bands and the Republican-win tail together.
- A Dominant National Event Before November 3. A Dobbs-style ruling, a security crisis or a major scandal can reset a midterm's whole frame in weeks, as 2022 showed when a June decision erased an expected red wave. It could push the margin either way, but it is the fastest route out of the current stable picture.
- Evidence Of A 2026 Polling Bias. The graded-poll archive the models used stops at 2023, so 2026 pollster accuracy is unmeasured. A clear sign that this cycle's polls are missing in one direction by more than about three points would pull the panel's true center with it.
A Worked Example: How A Band Pays
The mechanics, not a recommendation. Every Kalshi contract settles at exactly $1 if its band contains the final margin and $0 if it does not, so a price in cents reads directly as an implied probability. At 8 cents, $8 buys 100 contracts of the Democrats +4 to +6 band. If the final national margin lands there, the 100 settle for $100, a $92 profit before Kalshi's trading fees. If the margin lands in any other band, the $8 is gone. There is no partial credit, and a margin of 6.001 points resolves in the next band up, not this one.
The panel's 20% blend on that band implies a fair price near 20 cents against the market's 8. The distance between those two numbers is this article's whole argument expressed on one contract, and you can run any price-against-probability pair through the free EV calculator to see what a gap that size is worth per dollar. Every number in this piece gets graded in public once the market settles.
Settlement Timeline
| Election Day | November 3, 2026 |
| When The Answer Is Known | The winning band is effectively set once final national House vote tallies are compiled, in the weeks after election night as states certify |
| Contract Expiration | Kalshi lists an expected expiration of November 3, 2027, a buffer for final certified totals |
| Next Catalysts | The generic-ballot aggregators refresh daily; presidential approval updates continuously; several states are still finalizing mid-decade redistricting maps that reshape individual races |
| Refresh Policy | This board is re-scored when the story moves, and every verdict on it grades in public at settlement. Prices as of August 19, 2026. |
To be explicit: these are model estimates for information and entertainment, not financial advice and not a recommendation to trade any contract. This analysis advocates for no party and no outcome. Kalshi lists CFTC-regulated event contracts for adults 18+ where legal.
The Bottom Line
The question people type into a search bar, who wins the House popular vote, has a boring answer this cycle that the market and five independent models reached the same way: Democrats, almost certainly. The interesting question is the one underneath, and there the market and the panel truly disagree. The market is priced for a Democratic margin that grows into a wave, with a 60% chance of eight points or more. The panel, working only from a generic ballot that has barely moved and a base rate the incumbent's approval leaves fully intact, sees a clear win that stays close to where the polling already is, and puts that same eight-or-more outcome at 33%.
One of those reads will be closer, and on November 3 it will be obvious which, on a scoreboard we keep in public either way.
