History of Prediction Markets

Modern prediction market platforms are young; the mechanism they run on is centuries old. People have organized markets on public events since at least the 1500s, and for long stretches the prices those markets produced were treated as forecasts. They were quoted in newspapers, argued over, and consulted by anyone who needed a read on what was likely to happen. Today's exchanges are the latest generation in a lineage that runs through Gilded Age Wall Street, a university economics department, a Pentagon research agency, an offshore exchange in Dublin, and a wave of blockchain protocols. Most earlier generations did not survive, and the ways they failed are as instructive as the ways they worked, because a small set of failure patterns repeats across eras that otherwise share almost nothing.

The full arc breaks into six overlapping eras:

Where did prediction markets begin?

The earliest well-documented markets on event outcomes formed around papal conclaves. By 1503, organized speculation on the identity of the next pope was established practice in Rome (Wikipedia). Elections, wars, and successions attracted similar activity across the centuries that followed, usually informally and usually at the edge of the law.

The thread that connects this early activity to modern markets is the treatment of prices as information. Where enough people back their expectations with money, the going rate on an outcome summarizes what they collectively believe, and observers can read that rate as a forecast. This is the property the modern field is built on. In the survey that defined the academic literature, economists Justin Wolfers and Eric Zitzewitz describe prediction markets as markets whose prices aggregate dispersed information into forecasts of uncertain future events (Wolfers & Zitzewitz, 2004). A conclave pool in 1503 was not designed for that purpose, yet it already displayed the essential behavior. Its price expressed a consensus expectation and moved as news arrived. How that forecasting purpose separates an exchange-traded event contract from a sportsbook line is covered in Markets vs Sportsbooks.

In the United States, the informal version attached itself to elections early. By the second half of the nineteenth century, it had organized into something recognizably market-shaped, with central meeting places, standing intermediaries, and prices the press reported as news.

Did prediction markets exist before opinion polls?

They did, and by decades. Between 1868 and 1940, organized markets on presidential elections operated in New York, centered on Wall Street, while the first scientific election polls did not appear until the mid-1930s. The economists Paul Rhode and Koleman Strumpf reconstructed this era from thousands of contemporary newspaper records in Historical Presidential Betting Markets, which remains the standard account.

The scale was substantial. Rhode and Strumpf document more than $165 million in 2002 dollars changing hands on a single presidential race, with activity that at times exceeded trading on the stock exchanges. In some election years, turnover in these markets amounted to more than half of total campaign spending (Wikipedia).

The press of the day frankly called this election betting, and the quoted odds functioned as the forecast of record, published the way later generations would publish polling averages. A quoted price on a candidate implied a probability of victory, the same relationship modern contracts make explicit (Prices and Probabilities walks through the conversion). By Rhode and Strumpf's assessment, the markets forecast outcomes well in the era before polling, and documented attempts to manipulate prices generally failed.

The era's end had identifiable causes. Rhode and Strumpf attribute the decline to the arrival of scientific polling, which gave newspapers a cheaper and more respectable source of election forecasts, and to the rise of other legal outlets for speculative money. By the Second World War, the New York election markets had faded, and for the next half century polls held the forecasting role that prices had held before them.

What are the Iowa Electronic Markets?

The modern, institutional form of the idea arrived in 1988 at the University of Iowa, where economists George Neumann, Robert Forsythe, and Forrest Nelson launched what was then called the Iowa Political Stock Market during that year's presidential race (Wikipedia). Participants traded contracts on the election outcome with real money, in small amounts, on a market run for research and teaching. The project became the Iowa Electronic Markets, or IEM, and it has operated continuously into the 2020s, the longest run of any real-money prediction market in the United States.

The IEM's legal design proved as influential as its research output. The Commodity Futures Trading Commission (CFTC), the US derivatives regulator, issued no-action letters stating that the agency would not recommend enforcement against the market, premised on its not-for-profit academic character and its small stakes, with individual accounts capped between $5 and $500 (Wikipedia). That research carve-out became the template other academic markets would follow, including PredictIt a quarter century later.

The IEM also standardized the contract designs the field still uses. Winner-take-all contracts pay $1 if a specified outcome occurs, so the price can be read as a probability; vote-share contracts pay in proportion to the final vote, so the price can be read as the market's expected vote share. Wolfers and Zitzewitz later formalized this taxonomy, mapping contract types to the statistic each price reveals (Wolfers & Zitzewitz, 2004).

The accuracy record is the reason the IEM appears in nearly every academic discussion of prediction markets. In Prediction Market Accuracy in the Long Run, Joyce Berg, Forrest Nelson, and Thomas Rietz compared IEM prices against 964 national polls across the presidential elections from 1988 through 2004 and found the market closer to the eventual outcome 74 percent of the time, with an average election-eve error around 1.33 percentage points and the advantage over polls largest months before the vote. A small academic market with capped accounts had produced, over five election cycles, one of the strongest forecasting records available, and that evidence anchored the academic case for the mechanism.

What was the Policy Analysis Market?

In the early 2000s, DARPA, the US Defense Department's research agency, funded a program called FutureMAP to test whether market prices could support intelligence analysis. Its flagship project was the Policy Analysis Market, or PAM, designed by a team that included economist Robin Hanson, whose archive of the original project documents remains the primary record of what was actually proposed. PAM would have listed contracts on geopolitical and economic conditions across the Middle East, on the theory that markets could aggregate dispersed judgments that intelligence bureaucracies struggle to combine.

The program never opened. On July 28, 2003, two senators held a press conference denouncing it as a taxpayer-funded "terrorism futures market" and calling the idea "ridiculous and grotesque"; the Pentagon canceled the program the following day, and the head of the office overseeing it resigned within weeks (Wikipedia). From denunciation to cancellation took roughly 48 hours.

Hanson's first-person account argues that the coverage misrepresented the design, and that most proposed markets concerned broad indicators such as economic and political stability rather than individual attacks. The episode nonetheless demonstrated that markets on sensitive events can be ended by controversy alone before any trading takes place, a lesson later platforms would relearn.

The cancellation had a second, less expected effect. The controversy put prediction markets on front pages, and academic interest grew rather than shrank. The Wolfers and Zitzewitz survey that defined the field opens with the PAM episode (Wolfers & Zitzewitz, 2004), and five years later, twenty-two economists, among them five Nobel laureates, signed a short policy forum in Science arguing that prediction markets were a promising research tool and that regulation should accommodate them (Arrow et al., 2008). The political system had rejected the instrument; the research community had largely endorsed it. That tension ran through the following two decades.

What happened to Intrade?

Intrade carried prediction markets into mainstream awareness. Founded in Dublin in 1999 within the Tradesports family of exchanges and driven for most of its life by chief executive John Delaney, it listed contracts on elections, awards, economic releases, and world events for a global user base (Wikipedia). Through the 2004, 2008, and 2012 US election cycles its prices became a fixture of political coverage, and near the peak in October and November 2012 the site drew more than 50 million page views a month (Wikipedia).

The collapse came on two separate threads, and the regulatory thread broke first. On November 26, 2012, the CFTC filed a civil complaint against Intrade and its parent, Trade Exchange Network, for offering off-exchange commodity options to US customers and for violating an earlier cease-and-desist order. Intrade closed its US accounts within a month (Wikipedia), cutting off the users who had made it famous.

The internal thread surfaced later. Delaney had died in May 2011 while attempting to summit Mount Everest, and examinations of the company's finances after his death uncovered irregularities connected to Delaney himself (Reason). On March 10, 2013, Intrade suspended all trading, citing financial irregularities, and never reopened (Wikipedia).

Which thread actually killed Intrade is still debated. Katherine Mangu-Ward made the contemporaneous case in Reason that the CFTC's suit, which in her reading cited no evidence of customer harm, ended the experiment before the internal problems were public; the audit record shows the company had serious problems of its own regardless of the enforcement action. What the episode settled in practice was the legal question. After Intrade, a venue serving US traders would need a research carve-out, a registered exchange, or no US customers at all, and the next decade's platforms sorted themselves along exactly those lines.

How did Polymarket and Kalshi reshape prediction markets?

After Intrade, the field split into two threads that developed in parallel for a decade: one rebuilding prediction markets outside the regulated system, on blockchains, and one working inside it, through registered exchanges and research carve-outs. Polymarket would come to lead the blockchain thread and Kalshi the regulated one.

The regulated thread was older than it looked. HedgeStreet, a CFTC-designated exchange listing small event-based contracts, had opened in 2004 and struggled to attract volume, and its designation later passed to Nadex (Wikipedia). Early regulated attempts shared that liquidity problem, and in 2012 regulators prohibited Nadex's proposed political-event contracts, keeping election markets off regulated venues for another decade (MarketsWiki). The academic carve-out continued in parallel. In 2014 the CFTC granted no-action relief to Victoria University of Wellington for a small not-for-profit political market on the IEM model, which launched as PredictIt with caps on trader counts and position sizes. PredictIt served as the reference US political market for eight years, until the CFTC withdrew the letter in 2022 and ordered the market wound down. A federal appeals court later found the withdrawal likely arbitrary and capricious, and the saga ended with PredictIt approved to operate as a regulated exchange (Wikipedia).

The crypto thread moved faster and broke more often. Augur, built by the Forecast Foundation from 2014, raised roughly $5.3 million in a 2015 token sale, one of the first major crowdsales on Ethereum, and launched in July 2018 to intense attention; within a month, daily active users had fallen from 265 to 37 (Wikipedia). Its deeper problem was resolution. Because anyone could create a market, traders learned to write deliberately ambiguous markets engineered to resolve as invalid while their creators profited, a design flaw documented in detail by Binance Research. Veil, a trading interface built on Augur, lasted six months in 2019; its founder's postmortem cited trying to do too much, onboarding that non-crypto users could not get through, and an untenable middle ground between full decentralization and regulatory compliance. Polymarket, founded in 2020 (Wikipedia), became the era's most prominent crypto-native venue. Ethereum founder Vitalik Buterin traded the 2020 election on Augur-based infrastructure, and his firsthand essay recorded both the promise and the frictions of the on-chain model: prices that stayed arguably mispriced for weeks because capital was locked up and smart-contract risk deterred arbitrage, alongside an oracle that nonetheless resolved a contested election correctly. The thread then met the same wall as its predecessors in January 2022, when the CFTC ordered Polymarket to pay a $1.4 million penalty for operating an unregistered facility and to wind down noncompliant markets.

The two threads converged in the mid-2020s. Kalshi, an exchange holding full CFTC designation, won a federal court ruling in October 2024 that cleared its election contracts for trading (Wikipedia; the full docket is public). Election markets, refused a place on regulated venues in 2012, had won one in court. Regulated exchanges and crypto-native venues now form the two main families of the modern landscape, a structural split covered in Venue Types, and the regulator's posture shifted with the era. The same agency that sued Intrade and Polymarket has since published its own educational explainer of prediction markets.

What does the history of prediction markets suggest?

Viewed end to end, the record reads less like a straight line of progress than like a cycle in which the mechanism keeps proving itself, platforms keep dying, and the deaths cluster into three patterns.

Regulatory collision is the oldest pattern, and it arrives in several forms. PAM fell to a political veto before launch. Box-office futures were legislated away. In June 2010 the CFTC approved two exchanges to list futures on film box-office receipts, and weeks later, after studio lobbying, the Dodd-Frank Act banned the product outright (The Ringer; Wikipedia). Intrade and Polymarket met enforcement actions. iPredict, an academic market run from New Zealand, closed after its government declined an anti-money-laundering exemption, a death by compliance cost rather than by prohibition (Wikipedia). PredictIt's wind-down order was an administrative reversal that a court later found likely unlawful. Much of the pattern traces to the single contested question of whether an event contract is a financial instrument or a gambling product. Each era has relitigated where that line falls.

Demand failure is quieter in the record, and it may be just as common. HedgeStreet struggled for volume inside a fully regulated wrapper, Augur's users left within a month of launch, and Veil folded in six months. In a mid-2024 essay for Works in Progress, Nick Whitaker and J. Zachary Mazlish argue that regulation has not been the binding constraint at all. Election markets have long been fully legal in the United Kingdom, yet the authors counted only around £12 million in play on that year's US presidential race, roughly the amount a single cricket match typically attracts (Works in Progress). Legality, on this argument, is necessary but not sufficient. A market also needs a reason for enough traders to show up, and the eras where volume arrived are the eras where the product found one.

Resolution is the third pattern and the least visible from outside. Someone has to decide what actually happened. Augur turned ambiguity itself into an attack surface (Binance Research), and Buterin's account of the 2020 election shows how much of a market's credibility rides on settling a contested outcome correctly (Buterin, 2021). The rules for deciding the outcome, and the trust those rules command, are the hardest parts of a market on events to get right, a difficulty each era has rediscovered.

Against that pattern of platform failure runs a consistent empirical thread. The IEM beat polls across five election cycles (Berg, Nelson & Rietz). When Google, Ford, and a third firm ran internal prediction markets, prices improved on the companies' official forecasts by up to a 25 percent reduction in mean squared error, even with small stakes and thin participation (Cowgill & Zitzewitz). The aggregation idea itself, examined in Wisdom of Crowds, has outlived every platform that carried it.

For a reader of modern markets, the history carries practical weight. Prices as forecasts have a record stretching back a century and a half, which is worth respecting. The failure modes have a record just as long, which is worth checking. Whether a market's resolution terms are unambiguous, whether enough capital is trading for the price to mean something, and whether the venue's legal footing matches the trader's jurisdiction are the same three questions the graveyard keeps asking.

Related guides

The eras above compress a large primary record, and the concepts the history keeps circling back to each have a guide of their own.