Part three of “The price of tomorrow,” a series on prediction and futures markets.
The first two parts of this series established what prediction markets are and what they demand of us ethically. Part one argued that while futures markets exist to transfer risk, prediction markets exist to aggregate belief. Part two warned that pricing uncertainty, especially geopolitical or catastrophic uncertainty, carries structural and moral risks that cannot be waved away.
In October 2024, Polymarket’s contract on the U.S. presidential election was pricing Donald Trump’s chances at roughly 60 to 66 percent, while poll-based forecasters such as FiveThirtyEight and The Economist’s own model had the race closer to a coin flip. Two informational systems, observing the same electorate, produced materially different numbers. The question this raises is not whether prediction markets are useful in the abstract. It is more specific than that: when a prediction market and a conventional forecasting method disagree by a wide and persistent margin, what, if anything, does that divergence tell an investor once the market’s own mechanics are examined?
The divergence as evidence, not just as theory
Start with what actually happened, because the mechanism only matters if it produced something real. Through most of October 2024, Polymarket’s Trump contract traded well above the probability implied by polling aggregators, and the gap widened rather than narrowed as Election Day approached. At the same time, the currency and rate markets most exposed to a Trump victory, the Mexican peso and short-dated U.S. Treasury yields among them, were not pricing that outcome with anything like the same conviction. The peso weakened modestly into early November. It did not weaken the way it would have if institutional foreign exchange desks shared Polymarket’s confidence.
That gap is the interesting fact. A prediction market and a rates desk were looking at the same election and arriving at different implied probabilities, and the size of that gap, not the level of either number on its own, is what should draw analytical attention. Futures and options markets carry balance-sheet constraints, hedging mandates, and margin costs that shape what gets priced and how aggressively. A pension fund hedging duration risk is managing its own book, not casting a pure vote on the election. A prediction market contract, by contrast, has no such constraint attached to it.
A skeptic well versed in market microstructure would argue that this framing gets the comparison backward: prediction markets are, almost by construction, thinner and less liquid than the currency and rates markets against which they are being measured, and thin markets are not more informationally efficient than deep ones simply because they lack institutional hedging constraints. On this view, a divergence between Polymarket and the peso is not evidence that the crowd knows something the professionals do not. It is evidence that a shallow, retail-accessible venue is more exposed to noise, sentiment, and a handful of large bets than a market with orders of magnitude more capital behind it, and treating that noise as signal simply because it comes from a “prediction” market rather than a “hedging” market is a category error dressed up as an information-aggregation theory. This is not a fringe position. It is the position most consistent with standard market-efficiency reasoning, and it has to be answered on the specific facts of this case, not dismissed as a failure to appreciate what prediction markets are for.
The answer, worked out below, is that the skeptic is right about the mechanism and only partly right about the conclusion. The Polymarket divergence in October 2024 was, in fact, distorted by exactly the kind of concentrated, thin-market dynamic the skeptic describes. But the distortion did not fully explain the divergence, which is a different and more specific finding than either “the market was noise” or “the market was signal.”
What the audit found, and what it did not resolve
A single French trader operating under the pseudonym “Théo” held positions across several accounts that, at their peak, accounted for a substantial share of the total volume on Polymarket’s Trump contract, reportedly in the tens of millions of dollars. That is concentration, not aggregation, and it is precisely the mechanism the skeptical view predicts: a market thin enough that one well-capitalized actor can move the posted price for everyone else is not demonstrating the wisdom of a crowd. It is demonstrating the conviction of an individual, expressed at a scale large enough to look like consensus.
Three questions determine how much weight a divergence like this one deserves. Is the market sufficiently liquid that a single participant cannot move it unilaterally? Is open interest concentrated in a small number of hands, or distributed across a genuinely broad participant base? And is the resulting price a product of dispersed information being aggregated, in the sense Hayek meant when he described markets as processing information no central planner could collect, or is it one confident position dressed up as a market consensus?
Applied to Polymarket in October 2024, the answers are not clean, and this is where the skeptic’s mechanism and the analysis converge. Théo’s positions were real and large enough to matter, and multiple outlets reported that his trading contributed meaningfully to the size of the Trump-side premium. But the premium did not evaporate when his positions were modeled out. Analysts who attempted to strip his volume from the book still found Polymarket pricing Trump higher than the polling aggregators did. Concentrated positioning explains part of the divergence, not all of it, and that residual is the fact the skeptical view, applied mechanically, would have discarded along with the noise it correctly identified.
This is the specific point at which a purely structural audit reaches its limit. The audit is right that the market was distorted. It cannot, on its own, tell an investor whether the undistorted residual reflects real information or a different, unmeasured source of noise. Structural soundness and predictive accuracy are not the same test. On Election Day, Trump won comfortably, and Polymarket’s directional lean, whatever combination of concentrated conviction and genuine information produced it, priced the outcome more accurately than the polling aggregators had. Whether that accuracy was earned by information the market aggregated or was a distorted signal that happened to point the right way is a question the audit identifies but does not answer.
What this leaves unresolved for execution
This is where the two positions this section opened with, that the divergence should be traded directly and that it should be dismissed as noise, both come apart under the specific facts of the case. The skeptic’s mechanism, thin markets are noisier than deep ones, was correct. The skeptic’s implied conclusion, that the noise fully accounts for the divergence and therefore the signal can be set aside, was not supported once the position was modeled out. An investor relying on the audit alone, in either direction, would have drawn a conclusion the evidence here does not fully license.
What the case does establish is a distinction rather than a rule: a flagged, partially explained divergence is a different object than either a clean signal or a discredited one, and it carries a correspondingly different weight in a portfolio, somewhere between full conviction and full disregard rather than at either pole. Frank Knight’s distinction between risk and true uncertainty describes why that middle position is the honest one rather than an evasion. Polymarket’s number implied a single, clean probability for an event that would happen exactly once. The audit does not convert that number into a stable, repeatable frequency, whether or not the audit comes back clean. It only describes how much of the number’s apparent precision survives scrutiny, and in October 2024 a meaningful part of it did, while a meaningful part of it did not.
Reading the smoke, not steering by the alarm
An investor who dismissed Polymarket entirely, on the reasonable structural grounds that thin, concentrated markets are not reliable information aggregators, would have been right about the mechanism and would still have missed a residual divergence that traditional foreign exchange and rates markets were slower to price. An investor who traded the contract directly, treating its headline number as a clean read of crowd belief, would have overstated what the audit could support, even though the market’s direction happened to prove correct.
Prediction markets are neither flawless oracle machines nor useless gambling venues; they are early-warning signal aggregators. An investor should keep track of them. Ignoring them means turning a blind eye to real-time, financially backed information that traditional markets often price too slowly. However, portfolio adjustments should never be automated. Treat prediction market data as a smoke detector rather than a navigation system. When the detector sounds, you do not steer the ship based on the alarm alone; you look out the window, check your instruments, audit the signal, and then adjust your course.
This concludes “The price of tomorrow,” a three-part series on prediction markets, futures, and the ethical and practical boundaries of pricing the future.
Disclaimer: This article is for informational purposes only and does not constitute investment advice. Investors should conduct their own due diligence or consult with a financial advisor before making any investment decisions.
Photo by Boxed Water Is Better on Unsplash.
