Our mission to help you navigate the new normal is fueled by subscribers. To enjoy unlimited access to our journalism, subscribe today.
You often hear political pundits claim that the contest for the White House is a remarkably stable one. It’s true that Joe Biden has enjoyed a consistent lead in the average of battleground polls that’s toggled for months in a relatively narrow range between 5% and his current edge of 3%, the same margin he held in early September. But the polls look backward, reflecting voter preferences that are often five days old, or even further out of date. The data that best reflect where the race stands in real time, minute by minute––the prices gamblers are paying on the political betting sites––show a battle that’s turned wildly volatile in the closing days.
The proof is the fluctuating fever chart for each of the three states that together constitute a must-win for Trump: North Carolina, Arizona, and Pennsylvania. The odds in the Tar Heel State careen from red to blue practically by the hour. In Pennsylvania and Arizona, where he’s never led, the red line is zigzagging from steep dips that should give the Democrats comfort, to spikes toward even money that should make them nerve-janglingly nervous.
“Over the past week, we’ve seen few if any events that have moved the market. Instead, the big theme has been a jump in bettors’ uncertainty about the outcome, in different states and overall,” says Thomas Miller, a data scientist at Northwestern University who harvests the data from the U.S. gaming venue PredictIt.org to project overall odds each hour, posted on Twitter @virtualtout. “We’re now seeing far and away the greatest volatility in the entire campaign.”
The gambling sites show big, sudden shifts, with Trump pulling closer
Miller’s model uses data from odds on PredictIt for all 56 jurisdictions that offer electoral votes. He then marshals machine learning to run 1 million hypothetical elections an hour using Monte Carlo simulations that incorporate a broad range of possible outcomes in each state. He distills all the number-crunching into two forecasts that move in tandem: The number of electoral votes Biden is most likely to win, and the Democratic candidate’s odds of victory. Miller posts those benchmarks each hour on Twitter @virtualtout, along with a graph that shows the trajectory of the votes and odds. Displayed on the curve are key events that are aligned with the ups and downs for each candidate, from Trump’s brutish performance in the first debate and COVID-19 diagnosis that depressed his chances, to Michelle Obama’s heartfelt video and the 60 Minutes interview that lifted Biden’s fortunes.
Miller prefers betting odds to polls for three main reasons. First, prices the gamblers are paying for each contender are posted instantaneously, so the odds are always up to the minute, versus surveys that reflect opinions that are several days old. Second, the bettors aren’t expressing an opinion on whom they’ll vote for, meaning whom they think should win, but putting “skin in the game” by risking their own cash. Third, the “sample size” on PredictIt is much larger than for any poll. The site has over 100,000 active U.S. investors across all political markets.
At midday on Nov. 2, @virtualtout put Trump’s odds of winning at 23%, versus 77% for Biden, rating the former Vice President a 3-to-1 favorite. But what’s gripping is how much that number has jumped around in the past few days. It’s helpful to review the undulations in the forecast since the start of September. From Sept. 1 to Oct. 12, Biden’s chances of victory consistently stood at over 90%, hitting 94% and 326 electoral votes at the peak. Then Trump surged back when he returned to barnstorming, looking vital and eager for combat. By Oct. 28, he was posting his best numbers of the entire campaign at 25.3%. In less than three weeks, his odds have swelled from 1 in 10 to 1 in 4.
Then the race turned hugely volatile. Trump’s attack on Lesley Stahl on 60 Minutes appeared to rile female voters. By Oct. 30, just two days after hitting his summit, his @virtualtout number had shrunk by 11 points to 14.3%, or 1 chance in 7. But in 24 hours, Trump staged a stunning reversal, hitting 23.5% at 11:20 p.m. on Saturday, Oct. 31. Six hours later, late on Sunday afternoon, he’d fallen back to 18.3%. Then Trump roared back once again to 23% Monday morning, a gain of 4.7 points, before falling back to 22.8% by early afternoon. Miller puts his electoral vote count at 241, 56 shy of Biden’s projected 297. Hence, Trump’s chance has improved from less than 1 in 5 to, at times, around 1 in 4 in less than 24 hours.
Three states account for the wide swings in Trump’s odds
The most cited source for election forecasts, RealClearPolitics, lists six states as Top Battlegrounds: Florida, North Carolina, Pennsylvania, Arizona, Michigan, and Wisconsin. Biden leads in the average of polls in all but North Carolina, and is also narrowly ahead in a seventh, often labeled a toss-up, Georgia. PredictIt, however, gives a much more up-to-date picture of the prizes truly in play. Trump is now a big favorite to win Florida and Georgia, standing at 61% in both. The gamblers are pretty much counting him out in the upper Midwest; his odds in Michigan, Wisconsin, and Minnesota are all between 21% and 24%.
So the end game centers on the Big Three, North Carolina, Arizona, and Pennsylvania. It’s the oscillating odds in that triumvirate that account for the quicksilver shifts in Trump’s shot at victory. No state epitomizes the shifting currents better than North Carolina. From Oct. 1 to Oct. 18, Biden enjoyed a big lead, averaging almost 10 points. Then Trump came back, gaining a tiny advantage on Oct. 22. The lead changed hands three times in the next three days, but by Oct. 30, it looked like Biden was back in charge to stay when his edge jumped to 56% versus Trump’s 46%. That descent probably helped sink Trump’s odds at the time, since he has no path to winning without taking the Tar Heel State. But in yet another twist, on Sunday afternoon Trump grabbed the lead once again, and on Monday, has held a consistent margin of around 55% to 45%. It’s that switch that’s kept Trump in the game. As of today, he’s a narrow favorite to repeat his 2016 win in North Carolina.
Still, the whipsawing odds there are greatly adding to the weeklong surge in uncertainty. Another nail-biter is Arizona. It’s traditionally Republican territory that Trump took by 3.5 points in 2016. Yet on Oct. 8, Trump’s chances stood at just 32%, or 1 in 3. Although he was climbing back, on the Friday before the election he was in a hole at 43%. At midafternoon on Monday, he’s at between 47% and 48%, but he’s still never reached 50%. So Biden is maintaining a minuscule edge in the Grand Canyon State, though his prospects now stand on a knife’s edge.
So Trump holds what looks like a small but durable advantage in North Carolina and clings just below even odds in Arizona. The main reason he’s still such a long shot is the deficit in Pennsylvania. To be sure, Trump has rebounded strongly, doubtless aided by Biden’s gaffe on ending America’s oil and gas production, industries that account for 1 million jobs in the Keystone State. As recently as Oct. 7, Trump was lagging 27% to 73%. He clawed back to 43% on Oct. 27 and fell to 38% over the next three days. By midafternoon on Monday, he’d reached 42%, but had made no progress over the previous 24 hours.
Trump’s problem: Little time, not great momentum
And that’s precisely Trump’s towering challenge. He needs to hold North Carolina, where he’s just over even-money to win, overcome a small deficit in Arizona, and climb back from a shortfall in the odds of 58% to 42% in Pennsylvania. He’s got good to pretty good chances in each state. But to win, he’s got to pull off not one, not two, but a near-miraculous trifecta.
It’s important to address a yawning divergence in the overall odds for the nationwide, winner-take-all market on PredictIt, as well as the other betting sites, to those awarded by the Miller model. As noted, Miller’s methodology gives Trump a 1-in-4 chance of prevailing. But on PredictIt, Trump’s line on Monday afternoon is 40% to 60%, or 2 in 5 to win, 15 points higher than Miller’s approximate 25%. So if Miller is using the PredictIt data, why is the probability his model puts on a Trump victory so much lower? “It’s that people who are betting on the nationwide winner-take-all poll are using their feelings about what will happen in the election,” he says. “But to understand the outcome in the Electoral College, you have to model it. And that requires a computer.”
In fact, Miller is capitalizing on what he sees as a fun opportunity for arbitrage. He’s wagered $50 on the front-runner, a bet that would return him around $75, or 50% on his bet. “When I can buy something for $60 that my model shows is worth $75 to $80, I’ll buy that ticket,” he says. Miller has brought rigorous science to a field where predictions are often based on trends from the past and outdated polls. The question is whether any model can be smart enough to capture what Trump will need, a sudden surge in momentum that could put him over the top. “The time for that to happen is so short,” says Miller. The Big M is on Trump’s side, but with only hours to go, it doesn’t look strong enough to score the trifecta.