#A RACE THAT ISN'T A RACE
Polymarket took in something like $3.5 billion on the 2024 U.S. presidential race, and in the final days its price implied roughly 58-42 odds on Trump over Harris. The Electoral College came in 312-226. A lot of people looked at that and concluded markets beat polls, since polls in the same window were showing something closer to a toss-up. That's the wrong lesson, or at least an imprecise one.
The framing assumes markets and polls are two competitors answering the same question, where only one can be right. Christopher Wlezien, who has studied both for decades, has called that framing 'kind of silly': a poll measures a stated preference at a specific moment, a market prices an eventual outcome. Score them against each other and you're grading a thermometer against a weather forecast.
#WHAT THE MARKET IS ACTUALLY PRICING
Here's the mechanical point that gets lost: a political contract on Polymarket or Kalshi doesn't get its edge from some information source polls lack. Most of what moves the price is the same polling data everyone else sees, repriced within minutes instead of sitting untouched until the next release. When Berg, Nelson and Rietz compared Iowa Electronic Markets vote-share forecasts to 964 polls across five presidential elections from 1988 to 2004, they found the market closer to the eventual result 74% of the time, and the market's edge grew the further out from election day you measured.
- ▸Average poll error across the sample: 3.37 percentage points
- ▸Average IEM market error: 1.82 percentage points
- ▸IEM error on election eve: 1.33 percentage points
#THE RIGHT TEST
That pattern is exactly what continuous absorption looks like, not independent insight. A poll is a snapshot published every few days; a market price updates every time someone trades on a new one. Of course a running composite of every poll plus fundamentals plus whatever else gets priced in tracks reality better than any single snapshot; that's not the market beating polls, that's the market being built partly out of polls, updated faster.
So the interesting question isn't who wins a head-to-head, it's what an ensemble loses when you delete either input. A 2026 NBER working paper (w33339) actually ran that test, building a model that combines polls, economic fundamentals, market prices, and (crucially) the correlation between states, since a state that breaks one way tells you something about its neighbors. Ignore that cross-state correlation and a candidate's win probability can be biased by more than 10 percentage points. Include it, and the combined model tracks actual state-level outcomes better than the market price alone, state by state and in aggregate.
The Error Gap, From Early Polls to Election Eve
illustrativeMarket forecasts carried consistently lower average error than the polls feeding them, narrowing to a real 1.33-point gap by election eve; the shape of the curve between time buckets is an illustrative interpolation of the study's reported aggregate averages, not a published point-by-point series.
#WHERE THE MARKET ALONE BREAKS: DOWN-BALLOT
The presidential race is the one contract where absorption actually had fuel to work with: $3.5 billion in wagers means enough people were reading every poll, ad buy, and early-vote report and pricing it in near-continuously. Drop to a Senate or state legislative race and that fuel mostly isn't there: thinner books, wider spreads, a fraction of the attention. The mechanism that made the presidential price sharp doesn't have enough traders behind it to do the same job further down the ballot.
Polymarket called the top of the ticket right in 2024. It's a much shakier bet that the same contract type, at the volume down-ballot races actually draw, was outperforming a statistical model that pools fundamentals with the state-correlation structure the NBER paper found matters so much. A thin market reflects whoever bothered to trade it as much as it reflects the underlying odds; a model built to exploit correlated state outcomes doesn't need volume to do that. The headline race is the best case for market absorption, not the typical one, which is exactly why it's the wrong race to generalize from.
#PRICES AREN'T IMMUNE TO NOISE EITHER
None of this makes markets a lesser instrument, but absorption cuts both ways: the same channel that lets a market ingest a poll in real time also lets it ingest a manipulator with a large enough bankroll. The canonical case is Intrade in 2012, when a trader nicknamed the Romney Whale spent two weeks pushing the Romney contract higher and lost roughly $4 million doing it before the price snapped back. Polls have their own failure modes (nonresponse bias, herding toward the last released number), but they don't have that one. A market price is a function of who's willing to pay to move it, not a pure Bayesian update on public information, and treating it as the latter is its own category error.
#MY VERDICT
Stop scoring markets against polls like a horse race with two entrants. A market price is best understood as a fast, well-incentivized, occasionally manipulable way of aggregating polls, fundamentals, and news continuously instead of every few days, and the poll-and-fundamentals model is where the structure the market doesn't generate on its own, like cross-state correlation, actually lives. The only question worth asking before an election is which piece you'd delete first. On the top-line 2024 race, delete the market and you lose days of fast-moving signal you'd otherwise wait on a poll to confirm. On the other forty-some races on the ballot, delete the market and you probably lose almost nothing; delete the model and you lose the forecast entirely. Build the ensemble. The individual runners were never actually racing each other.
Where the Absorption Fuel Runs Out
illustrativeThe 2024 presidential contract pulled in a verified $3.5 billion on Polymarket, enough volume for real-time price absorption to work, while a typical down-ballot Senate or House contract (shown as an illustrative, order-of-magnitude estimate) doesn't come close, exactly where the market-alone story breaks down.