In March 2026, Kalshi — the largest regulated prediction market in the United States — reported $11.39 billion in monthly trading volume. Of that, $9.9 billion, or eighty-seven percent, came from sports event contracts. The figure is from Kalshi disclosures summarised in industry reporting and confirmed via Dune Analytics.
The second fact is that on February 28, 2026 — the day US and Israeli strikes began against Iran — Polymarket set a single-day volume record of $425 million, surpassing Election Day 2024. The volume was driven almost entirely by Iran-related markets resolving simultaneously. Roughly four days earlier, the Trump administration had issued a staff-wide email warning against placing bets on those Iran-war markets.
These two facts sit against a third. Mainstream news coverage — the Financial Times, the Wall Street Journal, Bloomberg, the New York Times — now routinely cites prediction-market prices as if they were forecasts. Google Finance embeds Kalshi and Polymarket odds alongside stock quotes. A 60 Minutes segment on Polymarket’s CEO ran in February.
The case is that the prices being displayed, and increasingly relied on, are not doing the epistemic work they are being credited with.
WHAT PREDICTION MARKETS ARE SUPPOSED TO DO
The intellectual case for prediction markets is straightforward and has been made, at various levels of rigour, since the Iowa Electronic Markets work of the 1990s. A market in which participants bet on future events using their own capital will, under specific conditions, aggregate dispersed private information into a price that is better calibrated than any single forecaster. The conditions are: adequate liquidity, diverse participants with diverse information, clear resolution criteria, and the absence of systematic bias — including the bias introduced when some participants have private information that others structurally cannot access.
When these conditions obtain, prediction markets are an excellent forecasting instrument. The meta-analytic case — Wolfers and Zitzewitz (2004), Arrow et al. (2008), Berg et al. (2008) — established the basic point. The Iowa Electronic Markets tracked presidential election outcomes, in aggregate, closer to the final result than major polling aggregators in several cycles. The academic case is not in dispute.
What is in dispute is whether contemporary prediction markets, at their current scale and composition, satisfy the conditions. The evidence from 2026 is not flattering.
WHAT THE VOLUME COMPOSITION SAYS
Eighty-seven percent of Kalshi’s March 2026 volume was sports. Polymarket had more than four thousand active sports markets as of April 7, 2026. The twenty-four-hour sports trading volume on Polymarket exceeded $120 million in March.
Sports markets are not epistemic infrastructure. They are a betting product regulated under a different framework — the Commodity Exchange Act, via the CFTC — from state-level sports betting regimes, for reasons that have to do with the Third Circuit Court of Appeals’ March 2026 ruling rather than with any substantive difference in what the product does. The fact that a prediction market can host a market on the NBA Finals does not say anything about its capacity to aggregate information about the Federal Reserve’s July decision. The two are different activities that happen to share infrastructure.
What the eighty-seven-percent figure establishes is that the industry’s revenue and volume base is sports betting. The non-sports markets — political, economic, geopolitical, scientific — are the minority. They are also the markets that carry the epistemic overclaim, because they are the markets that appear in newspaper citations and in policy discussion.
The industry’s commercial defence of this composition is that sports volume subsidises the lower-volume, higher-information markets — that a prediction market needs a liquid base to be useful on anything. This is plausible as a business model. It does not, however, make the lower-volume markets better calibrated. It makes them structurally dependent on the sports operation. If the sports operation is curtailed by regulators or competition, the epistemic product weakens with it.
WHAT THE INSIDER-TRADING PATTERN ACTUALLY SHOWS
In January 2026, a single Polymarket user reportedly took approximately $400,000 by correctly predicting the ouster of Nicolás Maduro. The event was not, at the time, publicly anticipated at the odds the trader faced. Subsequent Iran-war markets drew the Trump administration’s staff-wide warning in late February and early March.
A prediction market that is “more accurate than pollsters on the Maduro ouster” because a specific trader had private information about the ouster is not, in any functional sense, aggregating dispersed belief. It is laundering private information through a public price.
The concern, made explicit in congressional hearings in March 2026, is not that this or that trader had non-public information. It is that the structure of a prediction market does not differentiate between a trader with expensive public analysis and a trader with cheap private information. Both look, at the order-book level, like informed flow. Both move the price. A prediction market that is “more accurate than pollsters on the Maduro ouster” because a specific trader had private information about the ouster is not, in any functional sense, aggregating dispersed belief. It is laundering private information through a public price.
The structural vulnerability is worst on exactly the markets that carry the most epistemic weight. A market on whether Iran will close the Strait of Hormuz is most likely to be moved by someone close to Iranian or US decision-making. A market on whether a particular Federal Reserve decision will be a cut is most likely to be moved by someone close to Federal Reserve deliberations. A market on a particular regulatory announcement is most likely to be moved by someone inside the regulatory process. This is the paradox of high-stakes prediction markets: they are most useful where insider trading is hardest to prevent, and least useful on the questions where no one has private information worth acting on.
The specific pattern in 2025 and 2026 — Polymarket’s anonymous user base, the wash-trading estimates by outside analysts, the concentrated profits on politically high-stakes events — is what this structural vulnerability looks like when it materialises.
WHAT THE REGULATORY SITUATION ACTUALLY IS
Kalshi operates as a CFTC-regulated Designated Contract Market. Polymarket, after a 2026 acquisition of QCX, has partial CFTC recognition but continues to face state-level litigation and federal scrutiny. The Third Circuit, in March 2026, ruled that prediction markets cannot be regulated as sports gambling under state law. Nevada and Massachusetts have preliminary injunctions against Kalshi at the state level.
The practical effect is that prediction markets have, for the moment, won the jurisdictional question against state gambling regulators but remain exposed to federal oversight. The CFTC — as observers including the economist Stephan Bieri have noted — is, institutionally, the regulator that a regulatable entity would choose. Its resource base is modest; its enforcement record on market integrity in this product class is undeveloped; its mandate does not, cleanly, encompass the epistemic function prediction markets are now being credited with performing.
This is not a scandal. It is a predictable outcome of a young product routing into the weakest available regulatory surface. It does, however, mean that the integrity of the prediction-market price — the thing that newspapers now cite as if it were information — rests on a regulator that has not yet been equipped to verify it.
WHAT TO ACTUALLY SAY ABOUT A PREDICTION-MARKET PRICE
The defensible position is not that prediction markets are useless. They are not useless. They have demonstrated calibration advantages on specific classes of questions under specific conditions. The Iowa Electronic Markets, the Hollywood Stock Exchange in its active period, and various corporate-internal prediction markets all have documented track records.
The defensible position is that the prediction-market prices now being treated as information by mainstream media are, more often than not, failing the conditions under which those calibration advantages obtain. The volume is sports-driven. The headline political markets are structurally vulnerable to private-information flow. The regulatory guardrails against manipulation are undeveloped. And the citation practice — “Polymarket gives Hormuz closure at 34 percent, down from 52 percent yesterday” — treats a price that may have been moved by a single trader with knowledge of an administration decision as a read on collective belief.
A prediction-market price, in this environment, is a price. It is what people with money to bet thought was worth betting at. It is, in principle, an input into a forecast. It is not, in 2026, an output fit to be cited as a forecast without substantial additional work. That work includes: sourcing the liquidity, characterising the trader base, identifying the information advantage, and comparing the price to out-of-sample calibration data. None of that work is being done at the citation level.
THE DURABLE POINT
The durable question is whether prediction markets, as an institutional form, will develop the structural features that would make their prices actually forecastworthy, or whether they will remain, functionally, a sports-and-speculation business with a political sidecar that is useful mostly as content.
The answer will depend on regulatory choices that are being made now — at the CFTC, in state courts, in Congress — and on whether institutional participants with reputational stakes in calibration begin to dominate the marginal volume on political markets. Neither outcome is determined. The industry’s lobbying strategy — the mint-green Washington billboards, the Donald Trump Jr. advisorships, the Polymarket pop-up bar — is not, at present, oriented toward the kind of structural reform that would make the epistemic overclaim true. It is oriented toward preserving the commercial arrangement that produced it.
A serious reader, encountering a prediction-market price in the news, should do what a serious reader of any price does: ask who set it, what they knew, and whether the number corresponds to anything other than the demand for the contract. The answer, in many of the cases now being cited as consensus, is that it does not.
Prices are not forecasts. They are prices. The difference has been allowed to blur because the blur is commercially useful. It is worth, for once, unblurring it.
