At 6 a.m. EST on Election Day, The Virtual Tout, a website that forecasts political events, predicted that Donald Trump would be elected president of the United States for the second time.
The Virtual Tout predicted the Trump/JD Vance ticket would secure 280 electoral votes and the Kamala Harris/Tim Walz ticket would secure 258 electoral votes. Trump/Vance ended up with 312 electoral votes, the Harris/Walz team with 226 electoral votes.
Based on the underestimation of the Republican win, Tom Miller, owner and creator of The Virtual Tout, said, “I overcorrected from 2020.”
“There was a lot of uncertainty,” said Miller, who has a doctorate in psychometrics and runs The Virtual Tout’s election forecasting model. “The data shows it could have gone either way until the day before the election.”
Prediction markets more accurate than polls
The RoundTable spoke to Miller as a follow-up to the Oct. 22 article the RoundTable published about him and his work. He is also the faculty director of the graduate data science program at Northwestern.
On Election Day, 1.8 million shares traded on contracts for the two major parties in PredictIt, the market Miller relied on for the 2024 presidential election. (Other predictive markets, Kalshi and Polymarket, allow bets of $1 million, which skew the results so Miller does not rely on data from those markets.)
Miller explained how PredictIt operates in an email. “PredictIt provides the event contracts and serves as its own broker in selling contract shares, he wrote. “PredictIt charges fees to buyers of prediction market contracts.
“There is a fee for participation and a fee on profits. Traders are not betting against PredictIt. If you bet 50 cents on the Democratic ticket and the Democratic ticket wins, then you get $1 (less a small percentage of your 50-cent profit) after the election has been declared a Democratic victory. If the Democratic ticket loses the election, you lose the 50 cents that you bet. The same rules apply for bets on the Republican ticket,” Miller wrote.
In a virtual event hosted by Northwestern on Jan. 14, Miller presented a data debrief about The Virtual Tout’s 2024 results. He said during this presentation that based on how bets were being placed up until the day before the election, he believes the election could have gone either way.
He reminded listeners all predictive markets show what bettors think will happen, not what they’d like to happen. They are more accurate than polls for predicting the future, he said. Predictive markets are based on the premise of self-interest. Bets, or contracts, are placed with the hope that money will grow.
Taking stock of results
After the election, Miller said he was exhausted from the onslaught of requests for interviews, which did not ease up after the election results. He also received hate mail from both sides, which is draining and stressful. He described the email as “nasty, profane and crazy.” The nasty ones came from Trump/Vance supporters, he said.
In 2020, Miller used state contracts, each with active bettors, to essentially run simulated elections every hour. By the time of the election, he had run a million simulations using those 56 Electoral College markets. The results turned out to be very accurate, save for the 12-point discrepancy which favored the Republicans.
PredictIt did not offer state contracts within the Electoral College in 2024 due to ongoing litigation. There are 56 state contracts because Nebraska has four independent markets and Maine has three independent markets within their states for the Electoral College. The remaining 48 states and the District of Columbia follow a winner take all method. Thus 4 + 3 + 48 + 1 = 56.
Legal issues
On August 2, 2022, a Commodity Futures Trading Commission (CFTC) order instructed PredictIt to close its markets by February 15, 2023. PredictIt appealed the order. While the appeal was working its way through the courts, PredictIt proactively decided not to offer state contracts.
Separately, in September 2023, the CFTC issued an order disapproving (i.e., rejecting) “congressional control political event contracts self-certified by Kalshi.”
On September 12, 2024, U.S. District Court for the District of Columbia ruled that the CFTC exceeded its statutory authority in an appeal brought by Kalshi. The law firm BakerHostetler posted an online summary of the decision, calling it “a watershed moment in the permissibility of election betting, which has long been debated as critics sound the alarm on the potential for election manipulation.”
‘2024 was all over the place’
Without the congressional contracts available in 2024, Miller needed to figure out an alternative. He said during the NU event, “Instead of a bottom up approach, we were working from the states and running hypothetical elections. What we did is we took the prediction market prices, and then used past Electoral College history and our knowledge from 2020 about the way prices work in prediction markets in order to come up with a forecast.”
“Compared to 2020, 2024 was all over the place,” Miller said “Much shorter election cycle, uncertainties about Harris/Walz. The fact that they had to, in a sense, introduce themselves to the voters. And there were changes during the campaign in the messaging and the emphasis on various, you know, issues. So a lot of volatility in 2024.”
Uncertainty and volatility until the end
Miller said he believes the election could have gone either way up until the day before the election. To prove how this statement is reflected in the data, Miller showed the NU audience a candlestick chart developed from The Virtual Tout data. It shows forecasted electoral votes instead of prices placed the morning of the day before the election. Miller said the logic is the same whether using prices or votes.
Miller wrote in an email, “Each candlestick plot shows the forecasted electoral votes for a 15-minute period. For the candle itself, a clear rectangle or candle represents a 15-minute period for which the forecasted Democratic electoral votes were higher at the end of the period than at the beginning of the period (movement toward the Democratic ticket). A red candle represents a 15-minute period for which the forecasted Democratic votes were lower at the end of the period than at the beginning of the period (movement toward the Republican ticket).”
Lines that look like plus signs (+) or the letter T (right side up or upside down) show where the beginning forecast and ending forecast for that 15-minute period are the same. The ‘wick’ for each candle shows how the forecast changed over time.
Miller said the major takeaway from this chart is the large degree of uncertainty and volatility it shows. The traders were uncertain about who would win the Electoral College up until the day before Election Day.
Prediction markets as campaign resource
Miller said another reason prediction markets are important is as a campaign resource. He cited huge dips in Republican bets on July 31, 2024 after Trump met with the National Association of Black Journalists conference in Chicago and falsely suggested Vice President Harris had misled voters about her race. That day was a turning point on The Virtual Tout. Miller said, “That’s when it went from Republican to Democratic.”
Another big Republican dropoff was on Oct. 27 when during a Trump rally at Madison Square Garden. Among other things, he referred to Harris as “low IQ.”
The Harris/Walz campaign suffered a drop on Oct. 8, after Harris went on The View and said she would not have done anything differently than what President Joe Biden did.
Miller said he also believes if the Harris campaign had been studying the prediction markets, they would have seen the negative trends after Liz Cheney received a prominent role in the campaign (Oct. 3).
One factor with prediction markets is the inability to isolate different voter segments. Results are in aggregate. Miller said they know that most participants in prediction markets tend to be male and younger. They take more risks. He said, “We do know that at least in some previous experience it looked like they had a Republican bias.”
More refinements in store for 2028
Miller anticipates using tools from time series analysis and quantitative finance to refine his calculations.
During the online NU data debrief, he couldn’t say how often he’d be updating data in 2028. Given the relaxed regulatory market, he anticipates new markets to emerge. Miller said he will make his choices known before things heat up in 2028.
NU prof says betting data shows presidential election ‘could have gone either way’ until very end is from Evanston RoundTable, Evanston's most trusted source for unbiased, in-depth journalism.