StuPredicts

I've spent an unreasonable number of hours on this midterms piece. I now see probabilities when I close my eyes.

The 98% accuracy headlines were impressive. What happened behind them was even more interesting. x.com/i/article/2118…

LET’S GOOOO!

The US midterms are done and prediction markets X has been an absolute goldmine this week.

Big wins, some painful misses, traders celebrating, and people still fighting over whether prediction markets actually beat the polls.

• Joe Weisenthal (@TheStalwart): Absolutely loved election night. Says prediction markets did exactly what they were supposed to do, making confident calls before the election and updating ridiculously quickly as votes came in.

• Nate Silver (@NateSilver538): Not having any of the victory-lap nonsense. Argues that the best polling models were ahead of the markets, and that traders were largely following information the forecasters had already produced.

• G. Elliott Morris (@gelliottmorris): Has some receipts of his own. His FiftyPlusOne model gave Democrats a 98% chance of taking the House and 79% for the Senate. Says it was ahead of prediction markets on the big outcomes.

• Luana Lopes Lara (@luanalopeslara): Points out that the NYT only launched its election needle at 8:37pm, when Kalshi was already pricing a Democratic House at 99% and Senate at 98%.

• Luke Kawa (@LJKawa): Probably my favourite description of the whole evening. Prediction markets are sales, TV is quality assurance.

• Andy Masley (@AndyMasley): Watching the election at a party and wondering why all the traditional election coverage feels so slow compared with prediction markets.

• Justin Slaughter (@JBSDC): Pushes back on claims the markets had a disastrous night. Points to Kalshi’s figures showing favourites winning 479 of 487 resolved races and explains how he was following the markets to understand the developing Democratic wave.

• Arin Dube (@arindube): Makes a useful point amid the celebrations. Yes, markets did well at picking winners. But they, along with other forecasters, underestimated how big the Democratic wave would be.

• Jeremiah Johnson (@JeremiahDJohns): Gets stuck into the Brier scores and reckons The Argument/Split Ticket could finish with the most accurate election model, depending on the remaining results.

• @JamesSurowiecki: Notices something quite strange on Election Day. Despite all the noise, Senate market prices barely moved for long stretches. Iowa was one of the exceptions.

• @tracewoodgrains: Has a brilliant response to Kalshi comparing itself with Silver Bulletin and Split Ticket. If traders are reading those models and then performing about as well as them, surely that’s a pretty good advertisement for the models?

• @JoePostingg: Reckons prediction markets are especially useful once votes start coming in. Before that, they’re often following the same forecasts as everyone else. Once the counting starts, the speed becomes genuinely useful.

• Benjamin Freeman (@benwfreeman1): Challenges reporting suggesting election trading was stagnating. Says the comparisons weren’t giving Kalshi’s growth a fair hearing.

• Emily Nicolle (@emilyjnicolle): Looks at the other side of the volume story. Polymarket’s US-election trading before the midterms was much lower than ahead of the 2024 presidential vote, while Kalshi was broadly flat. Interesting when everyone is celebrating record activity.

• @MarshallCohen: Shares five takeaways from the election, including what market watchers saw regarding suspicious trading and why the Democratic victory could make life more difficult for the industry in Washington.

• @DustinGouker: Puts the election volume into perspective. Kalshi reportedly traded $1.9bn across the midterm cycle. It did $4.2bn in sports in a single recent week, without parlays.

• Krown (@KrownCryptoCave): Questions how an apparent 75% favourite could suddenly become an underdog. Then gets a Community Note pointing out the candidate never actually reached those odds.

And that’s before getting into the actual money people made and lost!

I’m also putting together a much bigger write-up on what the midterms really told us about prediction markets.

🇺🇸 US MIDTERMS: KALSHI VS POLYMARKET. WHO GOT IT RIGHT?

With election results coming in, I’ve been checking how the two prediction markets performed.

And crucially, how early they got it right.

Here’s the scoreboard across 14 selected outcomes that have been called, using market prices from November 2, the day before the election.

📊 THE SCOREBOARD

@Kalshi: 12–2

✅ 12 correct favourites
❌ 2 incorrect favourites

@Polymarket: 11–3

✅ 11 correct favourites
❌ 3 incorrect favourites

🏛️ BREAKDOWN

Senate: Kalshi 4–1 | Polymarket 3–2

Governors: Kalshi 6–1 | Polymarket 6–1

House district: Both 1–0

Senate control: Both 1–0

❌ WHERE DID THEY GO WRONG?

Two significant misses for both platforms:

South Carolina Senate: Annie Andrews won despite being priced at…

• Kalshi: 32.5%

• Polymarket: 31.5%

Texas Governor: Gina Hinojosa won despite being priced at…

• Kalshi: 37.5%

• Polymarket: 39.5%

⚖️ WHY DID POLYMARKET HAVE AN EXTRA MISS?

Iowa Senate.

Kalshi gave eventual winner Josh Turek a 51.5% chance.

Polymarket gave him 49.5%.

One technically got the favourite right, the other didn’t.

But with prices that close, I wouldn’t award Kalshi much extra credit.

⏱️ THE BIG QUESTION: WHEN DID THEY GET IT RIGHT?

Here’s how the scoreboard would have looked six days before the election, on October 28, using the same 14 eventual outcomes.

Kalshi: 9–5

Polymarket: 9–5

Some forecasts had been pointing in the right direction well before then.

Democratic Senate control, October 28:

• Kalshi: 64.5%

• Polymarket: 66.5%

Texas Senate, James Talarico:

• Kalshi: 66.5%

• Polymarket: 68.5%

⏳ AND THE HOUSE?

This could be one of the most interesting early calls.

Democratic House control, October 28:

• Kalshi: 90.9%

• Polymarket: 92.5%

By November 2:

• Kalshi: 93.8%

• Polymarket: 95.5%

We’re still awaiting an AP call, so House control isn’t included in the scoreboard.

🎯 MY EARLY TAKEAWAY

Kalshi 12–2. Polymarket 11–3.

Very similar results, with the only difference being a virtual coin flip in Iowa.

But a correct favourite days before an election deserves more credit than one that only emerges hours before the result.

And even a 65% favourite should lose sometimes.

The really interesting comparison will be how both platforms performed at 30 days, seven days and 24 hours before the election.

That’s where we’ll get a clearer picture of how useful these markets actually were as forecasting tools.

🇺🇸 TEXAS GOVERNOR: $114.7 MILLION TRADED. ABBOTT WAS STILL THE FAVOURITE. HINOJOSA WON.

AP has called the race for @GinaHinojosaTX. With roughly 83% counted, she leads @GregAbbott_TX 52.9% to 45.7%.

The prediction-market story is extraordinary.

📉 THE ODDS REVERSAL

• Mid-July: Kalshi gave Hinojosa just 8%.

• October 30: @TheStalwart flagged her rise to 39% on Polymarket.

• November 2: She reached 40% on Kalshi.

• Election day: Back down to 34%. Abbott was at 66%.

💰 THE VOLUME

• Kalshi: $107,383,792

• Polymarket: $7,363,267

• Combined displayed volume: $114,747,059

And there were some enormous trades on the other side of the result.

🎯 TWO $500,000 ABBOTT BETS

@MarshallCohen reported two separate, approximately $500,000 institutional trades backing Abbott on Kalshi, flagged by @betint_ai.

The later trade came at 64¢.

That’s roughly $1 million in Abbott exposure across two block trades. Whether they were held to settlement, or were hedges against other investments, isn’t known.

@jaketapper even asked Hinojosa about the earlier Kalshi trade on CNN.

Her response: “Some people have more money than cents.”

🐋 POLYMARKET WALLETS ON BOTH SIDES

@PolyInsider_ recorded:

• surf: $15,795 buying Abbott NO at 36¢. Potential profit if held: $28,080.

• Treadmilled: $5,260 buying Abbott NO at 36¢. Potential profit: $9,350.

• Arbguy: $32,000 buying Abbott YES at 64¢.

Meanwhile, Joe Weisenthal (@TheStalwart) told @tbpn before the vote:

“When a market has a Democrat at a 40% chance of winning Texas, my view is you got to take that seriously.”

📊 THE TEXAS DOUBLE FLIP

Kalshi also offered a Texas Governor–Senate combination market.

Before the election:

• Abbott wins governor, Talarico wins Senate: 46%.

• Hinojosa wins governor, Talarico wins Senate: 35%.

The second combination was the eventual outcome.

A market can price an upset as possible while still favouring the other side.

The more interesting question now is what happened to those institutional Abbott positions, and who took the other side.

🇺🇸 IOWA SENATE: AP CALLS IT FOR JOSH TUREK. A DEMOCRATIC FLIP IN A RACE THE PREDICTION MARKETS PRICED AT 50-50.

@turek4iowa (D) defeats @hinsonashley (R), flipping the seat held by retiring Republican Joni Ernst and helping Democrats secure control of the Senate.

AP called the race at 11:35pm ET.

With 93% counted:

🔵 Turek: 50.7%
🔴 Hinson: 46.2%

Some remarkable context.

Donald Trump won Iowa by 13.2 percentage points in 2024. Democrats hadn’t won a Senate election here since 2008.

Turek, a two-time Paralympic basketball gold medallist, has become the first Democrat since Barack Obama in 2012 to win a federal election in Iowa.

📉 HOW THE PREDICTION MARKETS MOVED

The reversal over the final days was extraordinary.

• December 2024: Kalshi’s Republican contract opened above 80%, before Hinson became the candidate.

• October 27: Hinson was still at 60% on Kalshi.

• November 1: Polymarket had Hinson at approximately 56%.

• November 2: Turek surged to 52% on Kalshi, taking his first lead.

• Election Day: Kalshi and Polymarket were both around 50-50.

After the result, Turek’s price shot towards 100%.

📊 WHAT DID THE POLLS SAY?

The final New York Times/Siena poll had:

🔵 Turek: 48%
🔴 Hinson: 47%

RealClearPolling’s final average was even closer:

• Turek: 47.1%
• Hinson: 47.1%

The polling average was tied. The prediction markets were effectively tied.

💰 OVER $22.6 MILLION IN REPORTED TRADING VOLUME

• Kalshi: $19,176,943
• Polymarket: $3,444,620

Combined: $22,621,563.

🎯 WHO BACKED TUREK BEFORE THE SURGE?

@PolyInsider_ recorded several interesting trades.

BreakTheBank

• Bought 136,352 Turek YES shares at 43.2¢.

• Total cost: approximately $58,900.

• Potential profit if held: $77,452.

But there’s more.

The same wallet also bought 155,819 Hinson NO shares at an average 43.3¢, costing approximately $67,522.

That position offered another $88,297 in potential profit.

Across those recorded trades, BreakTheBank could have made approximately $165,749 in combined profit if every position was held through a successful resolution.

zerosmart

• Bought 27,766 Turek YES shares at 42.8¢.

• Total cost: $11,884.

• Potential profit if held: $15,882.

📉 AND WHO WAS ON THE OTHER SIDE?

Another wallet, ndb1, took sizeable positions pointing towards a Hinson victory.

• $94,781 buying Hinson YES at 56.6¢.

• $104,717 buying Turek NO at 56.8¢.

That’s nearly $200,000 in recorded purchases backing the opposite outcome.

Before the election, former NBC political director @chucktodd identified an Iowa win as a potential indicator of a much larger Democratic Senate night.

That’s precisely the wider electoral significance the race eventually acquired.

📈 THE SEVEN-DAY REVERSAL

The chart from @betint_ai captures the story particularly well.

A Republican advantage narrowed. Turek moved into contention. The prices converged around 50-50.

Then the election result arrived and the market moved towards certainty.

What interests me is how much money was committed while the outcome was still so finely balanced, and how differently individual traders positioned themselves.

The polls said it was close.

The prediction markets gave both candidates a serious chance.

Iowa delivered a Democratic Senate flip that helped decide control of the chamber.