Polymarket Had the Midterm Winners at 98¢ Hours Before the AP Called Them. How Much Should That Count?

By 9 p.m. ET on 3 November 2026, Polymarket had Democrats at 98.95¢ or more to win the House, the Senate and the Texas Senate race. The AP called Texas just after 11 p.m. ET, the Senate shortly before midnight and the House at about 1:25 a.m. on 4 November. Take the odds seriously as a fast read of the count with money behind it. How much weight a Polymarket price deserves depends on how much money and how many holders stand behind it, on what the market's rules say will settle it, and on a track record that one election can't establish.

What did Polymarket show before the calls?

Democrats were favored in all three markets for the whole final week, and prices passed 98¢ hours before each AP call. Polymarket leaned the same way as Decision Desk HQ's model, only harder.

Hourly points from Polymarket's price history for the "Yes" share, in ET:

On the morning before the vote, Decision Desk HQ's model with The Hill gave Democrats 84% for the House, 65% for the Senate and 62% for Talarico; Polymarket stood at 94.5¢, 73.5¢ and 75.5¢. A Fox News poll of 1,398 likely Texas voters, taken 15 to 19 October, had Talarico and Paxton tied at 49%, while Polymarket priced Talarico at 61.5¢ to 63.5¢ over the same days.

The side the market favored won all three, and the market had put more weight on it than the model or the poll. That's three outcomes. A 75¢ price that comes in and a 62% forecast that comes in are both consistent with what happened, and telling which one was better calibrated takes hundreds of resolved races. Count this election in Polymarket's favor, as one data point.

Where does a Polymarket "chance" come from?

From the order book. The number on the page is the midpoint between the best bid and ask, set by people trading with each other. How much it's worth depends on the money, the spread and how many holders are behind it.

Polymarket's documentation says prices "emerge from supply and demand as users trade with each other" and that Polymarket doesn't set them. The displayed probability is the midpoint of the best bid and the best ask; if the spread is wider than 10 cents, the page shows the last traded price instead. A 75% on Polymarket is the middle of what buyers were bidding and sellers were asking at that moment; no model or survey sits behind it.

These three markets had real money behind that price. At 3:42 a.m. ET on 4 November, Polymarket's data put the House event at $63.0 million traded since it opened, $20.0 million of it in the previous 24 hours. The Senate event stood at $31.4 million ($13.0 million in 24 hours) and the Texas race at $21.8 million ($8.6 million). The spread on both chamber markets was 0.1¢.

Volume says how much changed hands, not who is holding. Polymarket's holder data, read at 3:41 a.m. ET, showed 5,134 wallets holding the Democratic "Yes" shares in the House market, and the ten largest held 39.8% of them. In the Senate market 5,833 wallets held the "Yes" shares, and the top ten held 34.3%. Those are snapshots taken after the calls, and a wallet isn't a person: one trader can run several.

The documentation also warns that the order book has no size limit and that large orders may move the price significantly. In a market with a few thousand dollars on the book, a few orders can move it several cents.

Before you trust a number, check three things: volume in the millions, a spread of about a cent, and how much of the leading side sits with a handful of wallets. The chamber markets passed the first two easily. On the third, ten wallets held a third or more of the Yes shares: a market can be deep and still rest on a few large holders.

Who decides the final answer?

Not the price. UMA's optimistic oracle settles each market under its rules text, after a proposal, a bond and a two-hour window for disputes.

A price never resolves a Polymarket market. According to Polymarket's documentation, someone proposes the outcome on UMA's oracle and posts a bond; the bond is $10,000 on each chamber market and $2,500 on the Texas markets, per Polymarket's market data. Anyone can dispute within two hours by posting the same amount. A first dispute starts a new proposal round; a second sends the question to UMA token holders, after a 24 to 48 hour debate and a vote of about 48 hours, 4 to 6 days in all. The vote can back the proposer, back the disputer, call the proposal too early, or return Unknown/50-50, which pays $0.50 a share. Polymarket's help center says it can't alter or reverse a resolution, and that outcomes are final once UMA finalizes them.

The rules text matters as much as the oracle. The Texas market resolves once the AP, Fox News and NBC all call the race for the same candidate. The chamber markets count voting seats, using official information and a consensus of credible reporting, and they can stay open as late as 4 January 2027 if independents haven't said which party they'll caucus with.

At 3:48 a.m. ET on 4 November, the Texas market had resolved for Talarico (at 2:11 a.m.) and the Republican Senate market had resolved No (at 2:48 a.m.). The Democratic Senate market had an outcome proposed, and in the House event only the Republican market had a proposal. Both Democratic markets were trading at 99.9¢ bid.

The odds and the payout run on two separate mechanisms. The price tells you what traders expect; the rules text and the oracle decide what you're paid. Read the rules under the order book before you read the price.

How accurate have Polymarket's odds been?

Polymarket's own figures are high, and independent work on 2024 found more mixed results. Read a 75¢ price as about three in four, which leaves room for a one-in-four miss.

Polymarket publishes an accuracy page, read through our server at 3:44 a.m. ET on 4 November. It counts how often the outcome leading at a fixed time before resolution was the one that happened: 98.6% four hours out, 96.6% a day out, 94.2% a week out and about 90% a month out, with a Brier score of 0.0599. The Brier score is the average squared gap between price and result; 0 is perfect, and always saying 50% scores 0.25. The page covers all resolved markets, gives no sample size or period, and shows that 78.3% of them resolved No.

That hit rate counts a question whose answer was clear for weeks the same as a close Senate race, so it says little about close races. Polymarket's own explanation for the No share is that there are usually more ways for something not to happen than to happen. The page does include a chart of expected against actual outcomes, which is the right test, and Polymarket's caption says the two "closely align."

Independent work looks different. Joshua Clinton of Vanderbilt and TzuFeng Huang studied more than 2,500 political markets on four platforms in the last five weeks of the 2024 presidential campaign. In the preprint, 93% of PredictIt's markets predicted outcomes better than chance, against 78% on Kalshi and 67% on Polymarket, and even the best-performing markets showed little evidence of efficiency. Clinton later told DL News that the paper's main point was not accuracy. Kalshi said the study misunderstood prediction markets and that calibration is the right measure.

So treat a 75¢ price as roughly three in four, which means a one-in-four miss is a normal outcome. And don't carry a platform-wide hit rate over to a close race, since that's where it says least.

Can big money move the odds?

For a while, if nobody takes the other side. Verified cases show one trader can hold a large position, volume can be inflated, and trades can come before public news.

In October 2024, four Polymarket accounts held about $28.6 million in positions on Trump winning the presidency, and CNBC reported a "stark shift" toward Trump on Polymarket that month. Polymarket said the accounts belonged to a French national with "extensive trading experience and a financial services background," taking a directional position based on personal views, and that it had not identified anything suggesting he manipulated or tried to manipulate the market. He agreed not to open further accounts without notice.

Volume itself can mislead. In November 2025, Columbia University researchers including Allen Sirolly posted a preprint that flagged about 25% of Polymarket's historical volume as likely wash trading, with a peak near 60% of weekly volume in December 2024. Decrypt reported the share at about 17% of all-time volume in election markets and 12% in politics markets. Sirolly said the team had no evidence the exchange itself was involved.

Trades can also come before public news. In January 2026, a newly created Polymarket account bet more than $30,000 that Nicolás Maduro would be out of office by the end of January, hours before the U.S. operation that captured him, and was paid about $400,000, according to Rep. Ritchie Torres. His bill to bar federal officials and staff from trading such contracts on nonpublic information had 43 co-sponsors by 14 April, Roll Call reported, and Sen. Elissa Slotkin introduced a Senate version. We found no report that either has passed.

A large bettor can keep a price where they want only while others decline to trade against them. In a market with millions in daily volume and a one-cent spread, that costs real money; in a thin one it can take a few thousand dollars. Since volume can be padded, check the holder count and spread alongside it, and when a price jumps without news, ask who is buying.

How can the odds affect politics?

Through the news. Major outlets now carry the odds, and one experiment found odds shift who voters expect to win. Nobody has measured an effect on votes.

The odds now reach people who never open a prediction market. Kalshi became CNN's official prediction-markets partner in December 2025, with its data integrated across CNN programming, and days later signed an exclusive multi-year deal to put its odds on CNBC from early 2026. In January 2026, Dow Jones agreed to show Polymarket data in The Wall Street Journal, Barron's, MarketWatch and Investor's Business Daily. Google Finance added Kalshi and Polymarket data in November 2025, BeInCrypto reported. Public figures quote the odds too. In October 2024, Elon Musk posted that Trump was leading "by 3% in betting markets," calling them "more accurate than polls" because money was on the line; The Block put Polymarket at 50.6% for Trump and 48.4% for Harris at the time.

The one test we found of what seeing the odds does is a survey experiment the Brennan Center for Justice published on 20 October 2026. It showed 1,994 registered voters pairs of hypothetical congressional candidates with randomly varied Kalshi odds and NBC News poll numbers. A candidate shown at 67% was 10 points more likely to be picked as the likely winner than one at 50%, and one at 84% was 20 points more likely, an effect similar to polls. A warning about prediction markets didn't reduce it. The study measured expectations of who wins, not votes.

Regulation has moved as well. In 2024 Polymarket did not take American traders, under a 2022 settlement with the CFTC, CNBC reported. Its site now says a separate U.S. platform, Polymarket US, is run by QCX LLC as a CFTC-regulated exchange, and that the international platform isn't regulated by the CFTC. In April 2026 the federal government sued Connecticut, Arizona and Illinois over their attempts to regulate Kalshi and Polymarket, the AP reported, and the CFTC chair said the agency would protect its exclusive authority over these markets.

Odds reach voters mainly through news outlets that now carry them. They can shift who people expect to win, and nobody has yet shown that they shift how people vote. When you see odds quoted in a broadcast, ask which market they come from and how much traded on it.

So how seriously should you take them?

As a price with conditions attached. Check the money and the holders, read the rules, compare with models and polls, and wait for the oracle.

Before you lean on an election price, on Polymarket or anywhere else:

1. Check volume and spread. Millions traded and a one-cent spread mean the price is hard to push; a few thousand dollars and a 10-cent gap mean it isn't.
2. Check concentration. If ten wallets hold a third of the leading side, a few decisions can move it.
3. Read the rules and the resolution source. The Texas market waited for three outlets; the chamber markets count seats and can wait for independents until January.
4. Compare with models and polls. When a price runs well ahead of both, as it did in Texas, ask what the buyers know or assume.
5. Remember that a price isn't a result. A market is settled when the oracle resolves it, and a 98¢ share still loses everything if the result goes the other way.

We read wallets the same way. Whalidate scores a wallet on its settled trades and on how much of its profit came from its best trade, as our methodology describes, because one big call says little about the next one. One correctly priced election is one data point for a market, too.

How did the whales in our directory trade the midterms?

Mostly on the Democratic side, and early. In the week before the vote they paid about 69¢ a share for the Senate and Texas and about 93¢ for the House.

We looked at buys from 26 October to 1 November by the 2,685 scored wallets in our public whale directory, whether or not they pass our checks. In each market, the Democratic side means the Democratic "Yes" share plus the Republican "No" share, since both pay on the same result.

Across the three markets, 197 wallets bought the Democratic side of at least one and 59 bought the Republican side; 31 did both. A share bought at 69¢ pays $1 if its market resolves that way, about 44% before fees on the Senate and Texas buys, against about 7% on the House buys at 93¢. The Texas market has already resolved for Talarico. We counted buys, and some of these wallets may have sold before resolution, so this isn't anyone's realized profit.

The Senate share was still at 75.5¢ when the first polls closed and reached 98.95¢ by 9 p.m. Most of the gain went to people who bought a week earlier at 69¢, and they took the bigger risk for it: at 69¢ the market still gave the other side about three chances in ten.

You can follow wallets from the directory on Whalidate. Pick a few, set your rules and run the strategy on paper first, which is free and needs no card. Live copying is on a paid plan. Whalidate takes no fee per trade and no share of profit; Polymarket's own trading fee applies in the markets where it charges one. Browse the whale directory or open the app.