According to noted data scientist Thomas Miller, that’s the unsteady state of the 2024 presidential race with twelve days to go.

For the past two months, this writer’s been closely following the forecasting from Miller, who’s a professor at Northwestern University. Miller’s calls proved dead accurate for both the 2020 presidential election, and the two Georgia Senate runoffs held two months later. In the former, Miller correctly foresaw that the contest was far closer than posited by the late polls, and tagged Biden’s victory within 12 electoral votes. In the Peach State contests pitting Republicans Kelly Loeffler and David Purdue respectively against Raphael Warnock and Jon Ossoff, the polls held a week before Election Day augured that both GOP candidates would win comfortably, in a twofer securing the party’s control of the upper chamber. Miller’s analysis showed the favorites heading for decisive defeats. The data scientist rang the bell, nailing the margin of victory for the pair of races combined within two tenths of a point.

In those Senate elections, Miller used forecasting tools that blended polling data and prediction markets. But he’s a strong proponent of the latter. This time, he’s exclusively deploying odds based on bettors place not on the candidate they plan to vote for, but the one they think will win. His data source is PredictIt, the most trustworthy platform for political wagering; the site handles giant average trading volumes of 39,000 shares a day. Miller adjusts PredictIt’s daily prices by applying his own methodology. For example, the preponderance of men on the site favor the GOP. Miller calculates the size of that bias, and tweaks the numbers to get what he deems the most accurate read.

Put simply, the Miller system finds that the PredicIt prices display a strong correlation to the popular vote split. His research also shows and that over all presidential elections since 1960, the percentage of nationwide ballots cast for each candidate translates closely to the number of electoral college votes (EVs) they receive. Each day, Miller runs the PredictIt prices through his construct to calculate the EV counts. At midnight, he posts the breakdown on his homepage, The Virtual Tout.

This writer rates Miller’s model among the best methods for gauging the election’s outcome for a simple reason: It banishes the noise from polls and pundits, and distills all the contradictory information out there into, at any one moment, into a single electoral vote count for each candidate set by people putting their own dollars on the line. His platform resembles the markets for stocks and bonds where all the investors’ disparate views get weighed, then expressed as one price, for say, Microsoft or the S&P 500.

This election’s been an dizzying roller coaster ride, and the candidate plunging in August and September rebounded big

The Virtual Tout result for October 22 was the latest in a recent series of shockers: It showed Trump leading by 154 electoral votes, 346 to 192. “In exactly one month, the forecast using the same model and PredictIt ‘investor pool’ underwent a complete reversal,” marvels Miller. On September 20, Harris appeared en route to an easy win. She claimed the 337 EVs to 201 for Trump. In the subsequent 32 days, the tally swung towards the former president by an astounding 252 EVs. The race became a mirror image of its status when Harris’ numbers peaked in the days following her excellent debate performance.

Even before the recent upheaval, each candidate looked poised for a landslide, once for Trump, and twice for Harris. The former president was crushing Joe Biden in the days following their debate. Then, after Biden’s withdrawal on July 21, the Vice President hovered for a week-and-a-half at well below 270 mark needed to win. The outlook changed dramatically after Trump appeared before the National Association of Black Journalists on July 21, and falsely claimed that his opponent misled voters about her race. That day, Harris vaulted ahead, and remained dominant for just over two months. She reached a first peak around the time of the Democratic National Convention in mid-August, then Trump gradually regained his footing, shrinking the gap to around a dozen EVs just before the debate on September 10. His weak performance at the face-off in Philly sent his numbers plummeting once again, and Harris commanded over 300 EVs from that day all the way through October 1.

It wasn’t until October 7, just over two weeks ago, that Trump gained the upper hand for the first time in more than two months, nudging 2 EVs in front. From there, it’s became a liftoff for the GOP standard-bearer. By October 11, the gap grew to 70, but dipped to just 42 EVs five days later. The tightening was short-lived. In less than a week, Trump’s bundle more than tripled to the 154 EV margin reached on October 22.

The Kelly charges gave Harris a big, sudden, sorely-needed boost

The very day Trump hit that 150-plus summit, General Kelly issued his condemnation of Trump as an erratic amateur unfit for the presidency. In a New York Times article published on October 22 that garnered explosive media coverage the next day, Kelly declared that Trump “falls under the general definition of fascism,” and “prefers the dictator approach to government.” According to the retired Marine Corps general, who served as the former president’s chief of staff for 17 months in 2017 and 2018, Trump ultimately “seeks the power to do anything he wants to do.” Kelly further asserted that Trump made positive comments to him about Adolf Hitler, a charge Kelly also made in a piece published the same day in The Atlantic.

Trump fired back, branding Kelly “a total degenerate” and “a lowlife” who invented the story “out of pure Trump Derangement Syndrome Hatred.” Kamala Harris pounced to capitalize on the Kelly pounding. “This is a window into who Donald Trump really is from the people who know him best,” the VP stated, adding that Kelly’s view proves once again that Trump is “increasingly unhinged and unstable.”

Kelly’s blast helped Harris’ odds on PredictIt, and as a result significantly improved her standing on Miller’s Virtual Tout. On Wednesday, October 23, Harris added 22 electoral votes, rising from 192 to 214, and Trump shrank by the same number Harris gained, by 22 from 346 to 324, a fall that shaved his lead from 154 to 110. Harris’ jump wasn’t as big as the 35 EV surge in her favor the day of the debate. But it was the first large, one-day increase she’s gotten since then, and for now, put her back on an upward slope following a month of sharp, virtually continuous decline.

For Miller, and other experts such as Allan Lichtman, the fundamentals contradict the data

For Miller, the question now is whether this change in the race’s “fundamentals”—Kelly’s sweeping denunciation and Trump’s name-calling in response—show up in what he believes best demonstrates where the election’s headed, the best “technicals” reflected in his electoral vote projections.

Miller draws a distinction between “technical” and “fundamental” analysis in predicting elections—and says the takeaways from the two systems now contradict one another. He stresses that these approaches apply in politics as well as financial markets, where they’re routinely deployed in handicapping price future action for stocks and bonds. In assessing securities, the “technicals” identify market trends and patterns that repeat over time, forming a roadmap for where prices are headed. The fundamental focus examines the underlying factors that historically determine the trajectory for a company’s shares or an index, including forecasts for the likes of profits, revenues, buybacks and R&D.

For elections, technical analysis involves crunching data collected from polling or betting sites, and refining the numbers to “scientifically” determine the odds each candidate will win. The deeply stat-dependent approach doesn’t consider such “fundamentals” as the combatants’ policies, personalities, or the economic circumstances at election time. For example, moving to the center, delivering a positive message of hope and inclusion, and displaying sterling character traits have long proven winning strategies.

Not this time. Following Trump’s remarkable surge, Miller perceives a big disconnect between the fundamentals and technicals. “I can’t explain what we’re seeing in terms of political wisdom or the basics of the way the two candidates have run their campaigns,” he told Fortune. He adds on the Virtual Tout site, “The Republican message has been a dark and anti-immigrant message, laced with disparaging comments about Harris. Trump vows to take revenge against his opponents if he wins the 2024 election.” By contrast, he finds that “The Democratic message has been hopeful and upbeat, offering unity rather than division.”

For Miller, Trump is a far-right extremist championing the most radical platform since the out-of-the-mainstream, ultra-conservative agenda that sank Barry Goldwater in 1964. Miller notes that overwhelming winner Lyndon Johnson occupied a center-left position similar to Harris’ moderate stance this year. Hence, following the fundamentals, Miller reckons that Harris should be winning, and winning big.

Miller highly respects the forecasts from the “oracle of American elections,” Allan Lichtman, history professor at American University. Since 1982, Lichtman has picked the victor in every presidential election, including Trump’s surprise win over Hillary Clinton’s in 2016. He argues powerfully that it’s the fundamentals not the data, that capture the true picture. The Lichtman template asserts that thirteen bedrock drivers or “keys” determine who hits the 270 EVs or above required to capture the keys to the White House. The challenger must check six or more boxes to prevail. The list includes: Whether the incumbent party’s candidate faced a tough challenge for the nomination, and if the economy’s currently in recession.

In Lichtman’s view, Harris scores on those two, and pockets seven additional keys. Trump gets only four. In He considers that the polls now showing the vice president in trouble are meaningless because they suffer from “margins of errors” of at least twelve points. “The polls should be consigned to flames,” he stated in a recent interview. Lichtman cites that the celebrated surveys, and such famed prognosticators as Nate Silver, proved radically wrong in calling for a Clinton win in 2016. Lichtman argues that in Trump versus Clinton, the pollsters far underestimated the voting strength for Republicans, and that this year, they’re missing the Democrats actual power at the ballot box. As evidence, he cites that the Dems far outperformed the pollsters’ predictions in the 2022 midterms and the special elections that followed. Lichtman says that Harris has a lock on the fundamentals, and whatever the polls and betting odds say, the fundamentals always prevail.

Miller agrees with Lichtman that the logic of past elections favors Harris—but he’s sticking with the data

Miller shares Lichtman’s concerns about polls. He’s also impressed by Lichtman’s reasoning and track record. But he raises counter arguments as well. The recent national polls, Miller says, are “catching up,” and reinforcing, the betting odds. As of October 23, the RealClear Politics average had Trump trailing by just 0.2%, compared to 2 points as recently as October 5. The Dems, Miller warns, would need a far bigger popular vote advantage on November 5 to swing the electoral college.

The data guy also focuses on that what he ranks as a powerful Lichtman key, the nation’s current economic condition. For Lichtman, that factor’s a big plus for Harris because “incumbents” get a substantial lift if we’re not in recession. But Miller says that while the macro numbers look good, Americans don’t feel good, so the situation that usually helps the party in the White House is now doing just the opposite, imposing a heavy drag for the VP. “The message from the Democratic Party is that GDP is growing strongly, unemployment is low, inflation is coming down,” he declares. “They spotlight all these good indices. But most people don’t think about GDP or that prices aren’t rising as fast as before. They think about how they have to work two jobs to get by, or that their grocery bills jumped hugely under Biden, and that they have no savings and because of high interest rates, can’t afford a mortgage to buy a first home or trade in the old car for a new one.”

So Miller deems that “not a recession” gets swamped by the dollar squeeze Americans feel in their own lives. He raises two other negatives for the VP. A big one is her recent avowal that she wouldn’t have changed any of President Biden’s policies. “Then how is she the candidate of change, as she claims?” asks Miller. He also observes that a huge voter contingent is tired of America’s support for foreign wars, and fear that we’ll be forced to send U.S. troops onto harms way. As a result, Trump’s isolationism is currently more appealing than Harris’ traditional, pro-NATO stance that advocates strong backing for Israel in battling Hamas, and Ukraine in its fight to defeat Russia and save its homeland.

Miller also rejects the idea that PredictIt’s prices are extremely unreliable, and should be ignored, because the bettors are mostly a male, pro-Trump cohort that wildly skews the odds towards the former president. “I keep getting these OMG emails from people saying ‘How can this possibly happen? It must be because the bettors are young and male and bet like they do on sports so they’re leaning Republican!’” He counters that that tilt is slight, and that his framework corrects for it. “Keep in mind that the same investors were saying just the opposite a month ago,” he declares. “If you believe the results on September 20 then you have to believe the numbers on October 22. You can’t discount one and not the other.”

So how does he assess the chances Harris can close the gulf over the remaining thirteen days, especially now that the Kelly onslaught is dominating the news, and already lifted Harris’ formerly falling EV numbers? “Based on one day’s data, we’re seeing a shift in the direction of the campaign,” says Miller. “The trading volumes on the prediction markets are increasing, indicating that more people are changing their minds and shifting to Harris. Additionally, new investors are likely entering the market.” He adds that Kelly’s disturbing characterization of Trump may help Harris frame a compelling closing argument. “It reinforces her message that the campaign’s not just about women’s rights but everyone’s rights,” he says. “Trump’s reaction to Kelly’s comments shows once again that he regards his opponents as enemies, and that his message is dark.” The Kelly scenario, he stresses, spotlights that Harris is winning on the fundamentals that usually decide elections, but so far aren’t resonating in 2024.

Miller cautions that we’ve seen huge moves in relatively short periods, and could witness still another earthquake by Election Day that swells support for the VP, notably if the Biden team engineered a sudden settlement ending the Israel-Hamas conflict or war in Ukraine. In conclusion, Miller states that “You can make a strong case for Harris on the fundamentals. Trump should not win, but the data still says he’s going to win.” Still, the former president’s position, he adds, isn’t as strong as on the day before John Kelly dropped the haymaker that could reboot the VP’s flagging fortunes.