What’s behind Trump’s surge in prediction markets?

Strange things can happen when you're in the doldrums.

I’ve always loved the word “doldrums” since I first encountered it as a kid reading The Phantom Tollbooth. In meteorology, the term refers to the Intertropical Converge Zone, a region near the Equator where prevailing winds battle to a draw, producing calm and monotonous weather. But the word's etymology precedes that usage, instead describing a state where you feel “down in the dumps” or even a little cranky out of boredom:

by 1803, "low spirits, the blues, the dumps," colloquial, probably from dulled, past participle of dull (v.) in the sense of "make (someone) slow-witted," with ending perhaps patterned on tantrum.

Well, we’ve entered the doldrums of the presidential campaign.

This summer produced rapid-fire action: Joe Biden dropping out amidst a massive amount of intraparty pressure, an assassination attempt (and later a second one) against Donald Trump, two debates, and the VP picks. But now, there isn’t any of that structure. After last week’s VP debate, no more major events are scheduled between now and November 5.

That doesn’t mean there isn’t news — there’s a lot, actually, between hurricanes and strikes and the continued escalation of hostilities in the Middle East. However, the political implications of this news aren’t so obvious.¹ Most people don’t vote based on foreign policy, for instance.

Meanwhile, the polling has also been rudderless — and there hasn’t been much of it lately. (We didn’t even run a model update on Sunday because no new polls met our standards.) Following her debate with Trump, Harris expanded her national lead from roughly 2 points to 3 points, and it’s remained there pretty much ever since. That extra point is useful on the margin since it might be just enough to help Harris overcome the Electoral College bias — but the race remains a toss-up for all intents and purposes.

And what will the final four weeks of the campaign look like? Well, maybe pretty much the same as this.

For our model, we calculate two types of error distributions for our popular vote forecast. (The error structure for our Electoral College forecast is quite a bit more complicated, but leave that aside for now.) First, there’s a parameter we call “drift”, which is essentially how much we expect the polls to shift between now and Election Day. Then there’s what we call “error” proper — basically, how wrong we expect the polls to be, as drawn from an empirical distribution of polling accuracy in all presidential elections since 1936. The thing is, we’re at the point where error is quite a bit bigger than drift. In other words, most of the uncertainty in the race now reflects the possibility of a polling error — not the chance that things will dramatically change between now and Nov. 5.²

Where do you see a little bit more movement? In prediction markets. At Polymarket, where I serve as an advisor, Trump’s odds are up to 53.6 percent, reversing a small Harris lead from a week ago.