"The best way to predict the future is to create a market for it." — Attributed to various prediction market researchers, paraphrasing Abraham Lincoln's apocryphal quote

Chapter 2: A Brief History of Prediction Markets

Chapter Overview

Prediction markets did not appear from nowhere. Their roots extend back centuries — to Roman merchants hedging grain shipments, medieval gamblers wagering on papal elections, and eighteenth-century insurers at Lloyd's Coffee House pooling collective judgment about maritime risk. What we now call "prediction markets" emerged from a long tradition of using financial incentives to aggregate dispersed information.

This chapter traces that history in full. We begin with the ancient and medieval antecedents that established the core insight: people with something at stake tend to reveal what they actually believe, not what they wish were true. We then follow the thread through the founding of the Iowa Electronic Markets in 1988 — the first modern academic prediction market — and through the explosive controversy surrounding DARPA's proposed Policy Analysis Market in 2003. We examine the rise and fall of InTrade, the quiet revolution of play-money platforms, and the regulatory battles that shaped what markets could and could not exist.

The story culminates in the modern era: Polymarket's rapid growth during the 2020 and 2024 U.S. elections, Kalshi's landmark approval by the Commodity Futures Trading Commission (CFTC), and the flowering of forecasting communities like Metaculus and Manifold Markets. Along the way, we meet the researchers — Robin Hanson, Justin Wolfers, Eric Zitzewitz, Kenneth Arrow, Charles Manski, and others — whose work gave prediction markets their theoretical foundations.

By the end of this chapter, you will understand not just what happened, but why it happened: the recurring patterns of innovation, resistance, regulation, and adaptation that have shaped prediction markets into what they are today. You will also build a Python visualization of the complete timeline, turning historical knowledge into an interactive tool.

What you will need: Basic familiarity with prediction market concepts from Chapter 1. For Section 2.10, you will need Python 3.8+ with matplotlib and pandas installed.

2.1 Ancient Roots: Betting and Collective Judgment

2.1.1 The Fundamental Insight

Long before anyone coined the term "prediction market," human societies discovered a powerful truth: when people wager their own resources on outcomes, the resulting prices contain remarkably good information about the future. This insight — that markets aggregate dispersed knowledge — is as old as commerce itself.

The mechanism is straightforward. Imagine a merchant in ancient Rome deciding whether to ship grain from Egypt to Ostia. He does not need to be a meteorologist, a naval engineer, or a political analyst. He needs only to observe what other merchants are willing to pay for grain futures, what insurers charge for maritime coverage, and what odds gamblers offer on the safe arrival of ships. Each of these prices encodes the private knowledge of dozens or hundreds of individuals — sailors who know the weather patterns, soldiers who know the political situation in Alexandria, shipwrights who know the condition of the fleet.

This is the essence of what Friedrich Hayek would later formalize as the "knowledge problem" in his 1945 paper "The Use of Knowledge in Society." Prices in markets convey information that no single individual possesses. Prediction markets are simply the purest expression of this principle, stripped of the complications of physical goods and supply chains.

2.1.2 Roman Grain Markets and Maritime Insurance

Ancient Rome operated sophisticated commodity markets that functioned, in part, as prediction mechanisms. The annona — the Roman grain supply system — depended on merchants making forward commitments about future deliveries. The prices of these commitments reflected collective expectations about harvests in Egypt and North Africa, the safety of shipping lanes, and the political stability of grain-producing provinces.

Roman merchants also developed early forms of maritime insurance, known as foenus nauticum (sea loans). Under this arrangement, a lender would advance money to a shipowner; if the ship arrived safely, the shipowner repaid the loan with substantial interest (often 20-30%). If the ship was lost, the debt was forgiven. The interest rate on these loans was, in effect, a market-derived probability estimate of shipwreck. Higher rates for journeys during storm season or through pirate-infested waters reflected the collective judgment of experienced traders about the likelihood of loss.

These were not prediction markets in the modern sense — they were not designed to elicit probabilistic forecasts about discrete events. But they contained the essential ingredients: financial incentives tied to future outcomes, with prices reflecting aggregated private information.

2.1.3 Medieval Papal Election Betting

The most direct ancestor of modern prediction markets may be the betting on papal elections that flourished in medieval and Renaissance Italy. Beginning at least as early as the fifteenth century, organized gambling on the outcome of papal conclaves was widespread in Italian city-states, particularly in Rome and Venice.

Historians have documented extensive papal election betting markets dating to at least 1503, when bets were placed on the conclave that elected Pope Julius II. These were not casual wagers among friends. They were organized markets with professional bookmakers, standardized contracts, and publicly observable odds. The Venetian ambassador to Rome regularly reported betting odds back to the Venetian government, which used them as intelligence about the likely direction of papal policy.

The historian Leighton Vaughan Williams has traced these markets back even further, finding evidence of papal election gambling in the fourteenth century. What makes these markets especially interesting is that they operated despite — or perhaps because of — the Church's official condemnation of gambling. The information they produced was too valuable to suppress entirely. Diplomats, merchants, and political leaders all consulted the betting odds as a supplement to their own intelligence networks.

The papal betting markets exhibit several features that would become characteristic of prediction markets:

1. Incentive alignment: Bettors risked their own money, encouraging honest assessment rather than wishful thinking.
2. Information aggregation: No single bettor had complete information about the deliberations of the cardinals, but the market incorporated signals from many sources.
3. Public prices: The odds were widely known and served as a common reference point for decision-making.
4. Liquidity: The markets attracted enough participants to produce stable, meaningful prices.

2.1.4 Early Horse Racing and Political Betting

By the seventeenth and eighteenth centuries, organized betting had become a fixture of European life, particularly in Britain. Horse racing, which had long been a sport of the aristocracy, developed sophisticated betting markets that refined the art of odds-making. The establishment of the Jockey Club in 1750 and the codification of racing rules created a stable framework within which betting markets could flourish.

Political betting followed naturally. In Britain, wagers on election outcomes were common by the eighteenth century, and by the nineteenth century, American political betting markets had become large, liquid, and surprisingly accurate. The historian Paul Rhode and economist Koleman Strumpf have documented extensive political betting in the United States from the 1860s through the 1940s. During this period, the New York Curb Exchange (later the American Stock Exchange) hosted active betting on presidential elections, with prices that closely tracked — and often anticipated — eventual outcomes.

Rhode and Strumpf's research, published in the Journal of Economic Perspectives in 2004, showed that these historical betting markets were remarkably well-calibrated. When the market gave a candidate a 70% chance of winning, that candidate won approximately 70% of the time. The markets also responded quickly to new information, adjusting prices within hours of major campaign events.

The decline of these markets in the mid-twentieth century was driven not by poor performance but by legal changes. The rise of scientific polling (beginning with George Gallup's successful prediction of the 1936 presidential election) provided a legal alternative for election forecasting, and the gradual expansion of anti-gambling laws made organized political betting increasingly risky.

2.1.5 Lloyd's of London: Insurance as Prediction

Lloyd's of London, founded in Edward Lloyd's coffee house in the 1680s, represents perhaps the most commercially significant application of collective prediction before the modern era. Lloyd's syndicate model — in which individual underwriters pool their assessments of risk and collectively set premiums — is, at its core, a prediction market.

When a Lloyd's underwriter sets a premium for insuring a cargo ship, she is making a probabilistic forecast: the premium reflects her estimate of the probability and magnitude of potential losses, combined with a margin for profit and uncertainty. When many underwriters compete to offer coverage, the resulting market price aggregates their individual assessments.

Lloyd's is particularly relevant to prediction market history because it demonstrated several principles that would later be formalized by prediction market researchers:

• Specialization and local knowledge: Individual underwriters specialized in specific types of risk (maritime, fire, life), and their premiums reflected deep domain expertise.
• Skin in the game: Underwriters personally bore the losses from bad predictions, creating powerful incentives for accuracy.
• Market discipline: Underwriters who consistently mispriced risk were driven out of the market, ensuring that surviving participants were well-calibrated.

The Lloyd's model also illustrated the limitations of prediction markets. Catastrophic, correlated risks — events that affect many policies simultaneously — proved difficult for the market to price accurately, a lesson that would be repeated in modern financial markets.

2.2 The Iowa Electronic Markets (1988–Present)

2.2.1 The Founding Story

The modern history of prediction markets begins in 1988, at the University of Iowa. Four professors in the Tippie College of Business — Robert Forsythe, Forrest Nelson, George Neumann, and Jack Wright — created the Iowa Electronic Markets (IEM) as an educational tool and research platform. Their original goal was modest: to demonstrate to their students how markets aggregate information, using the upcoming 1988 presidential election between George H.W. Bush and Michael Dukakis as a real-world laboratory.

The IEM was revolutionary in concept, if humble in scale. It was a real-money market — participants invested actual dollars, albeit with a $500 maximum — that traded contracts whose payoffs depended on the outcome of political events. The initial market offered two basic contracts: one that paid $1 if Bush won the popular vote, and another that paid $1 if Dukakis won. The market price of each contract, therefore, represented the market's estimate of each candidate's probability of winning.

The founders obtained a "no-action letter" from the Commodity Futures Trading Commission (CFTC), which stated that the agency would not take enforcement action against the IEM, provided it remained a small-scale academic research project with limited investment amounts. This letter — a regulatory instrument that would prove crucial to later prediction markets — established a precedent for operating prediction markets within the United States.

2.2.2 The Academic Purpose

The IEM was designed to answer a specific set of research questions:

1. Can small-scale markets produce accurate forecasts? The efficient market hypothesis, developed by Eugene Fama and others, suggested that markets with many participants and high liquidity could produce informationally efficient prices. But could a market with only a few hundred participants, investing small sums, do the same?

2. Do markets outperform polls? In 1988, scientific polling was the dominant method for forecasting elections. The IEM founders hypothesized that markets might do better, because they continuously incorporate new information and because participants with better information are naturally drawn to trade more.

3. How do markets respond to events? Debates, scandals, economic news, and campaign advertisements all affect election outcomes. The IEM provided a real-time window into how these events shifted collective expectations.

2.2.3 Track Record vs. Polls

The answer to the founders' questions exceeded their expectations. Over the course of more than three decades, the IEM has compiled a track record that has become one of the strongest arguments for prediction markets.

The most cited analysis of IEM performance comes from a 2004 paper by Joyce Berg, Robert Forsythe, Forrest Nelson, and Thomas Rietz, published in the American Economic Review. Analyzing data from IEM presidential election markets from 1988 through 2000, they found that:

• The IEM's election-eve prices predicted the two-party vote share with an average absolute error of approximately 1.5 percentage points.
• In 596 head-to-head comparisons between IEM predictions and contemporaneous polls, the IEM was closer to the final outcome 74% of the time.
• The IEM produced accurate forecasts even months before the election, when polls were still highly volatile.

These results were striking. The IEM outperformed not just individual polls but also polling averages, despite having far fewer participants and far less funding than major polling organizations.

Subsequent research has qualified these findings somewhat. In very close elections, the IEM's advantage over polls narrows, and there have been instances where the market exhibited systematic biases — particularly a tendency to overweight the chances of the favored candidate (a phenomenon known as the "favorite-longshot bias"). Nevertheless, the IEM's overall track record remains impressive and has been instrumental in building the case for prediction markets.

2.2.4 Key Research Papers from the IEM

The IEM generated a rich body of academic research that laid the groundwork for the broader prediction market literature:

2.2.5 Legacy and Current Status

The IEM continues to operate today, though its influence has been somewhat overshadowed by larger, more accessible platforms. It remains an important research tool and has expanded beyond elections to include markets on economic indicators (such as the Federal Reserve's interest rate decisions) and other events.

The IEM's greatest legacy is not its own performance but the research paradigm it established. By demonstrating that prediction markets could work — that small groups of motivated individuals, trading small amounts of real money, could produce forecasts rivaling or exceeding those of professional pollsters — the IEM opened the door for everything that followed.

2.3 The DARPA FutureMAP Controversy

2.3.1 The 2003 Proposal

In 2001, the Defense Advanced Research Projects Agency (DARPA) — the Pentagon's research arm, famous for funding the creation of the internet — began exploring the use of prediction markets for intelligence analysis. The project, known as FutureMAP (Futures Markets Applied to Prediction), was part of DARPA's Information Awareness Office, headed by Admiral John Poindexter.

The intellectual foundation for FutureMAP came from Robin Hanson, an economist at George Mason University who had been developing the theory of prediction markets since the early 1990s. Hanson argued that prediction markets could serve as a powerful tool for intelligence analysis, aggregating the judgments of analysts who might be reluctant to express dissenting views through traditional channels.

The concept was straightforward: create markets where intelligence analysts, military officers, and perhaps members of the public could trade contracts linked to geopolitical events — the stability of specific governments, the likelihood of weapons proliferation, economic developments in the Middle East, and similar questions relevant to national security. DARPA contracted with Net Exchange, a technology company, to build the platform, which was called the Policy Analysis Market (PAM).

2.3.2 Political Reaction

On July 28, 2003, Senators Ron Wyden (D-Oregon) and Byron Dorgan (D-North Dakota) held a press conference denouncing the Policy Analysis Market. Wyden described it as a "federal betting parlor on atrocities and terrorism" and argued that it was morally repugnant to allow people to profit from predicting terrorist attacks or assassinations.

The reaction was swift and overwhelming. Media coverage was almost uniformly negative, focusing on the most sensational aspects of the proposal — the possibility that someone might bet on when a foreign leader would be assassinated, or profit from advance knowledge of a terrorist attack. The nuances of information aggregation and the potential intelligence benefits were largely lost in the uproar.

Within 24 hours of the press conference, the Pentagon announced the cancellation of the Policy Analysis Market. Within a week, Admiral Poindexter resigned from DARPA. The entire project was dismantled.

2.3.3 What Actually Happened vs. What Was Reported

The gap between the actual FutureMAP proposal and its public characterization was enormous. Several important facts were lost in the controversy:

1. The market was not about terrorism directly. PAM was designed to trade contracts on broad geopolitical indicators — GDP growth in Jordan, military spending in Egypt, the stability of the Saudi monarchy — not on specific terrorist attacks. The idea that it would include "assassination futures" was a mischaracterization that the program's critics promoted and the media amplified.

2. The amounts involved were tiny. Like the IEM, PAM would have limited individual investments to small sums. It was a research project, not a Wall Street trading floor.

3. The intelligence community already used similar techniques. Intelligence analysts regularly make probabilistic assessments of geopolitical events. PAM would have formalized this process and made it more transparent, not created something fundamentally new.

4. The academic case was strong. By 2003, the evidence from the IEM and other prediction markets was substantial. PAM was a reasonable extension of proven methods to a new domain.

2.3.4 Lessons About Perception

The FutureMAP controversy teaches several enduring lessons about prediction markets:

Framing matters more than substance. The Policy Analysis Market was killed not by its content but by how it was described. "Betting on terrorism" is a devastating frame, even if inaccurate. Prediction market advocates learned — painfully — that public communication must anticipate and preempt hostile framings.

Political incentives can overwhelm policy logic. Senators Wyden and Dorgan were not making an analytical argument about the effectiveness of prediction markets. They were making a political argument about the optics of a Pentagon-sponsored betting market. The political incentive to oppose the project was overwhelming, regardless of its merits.

Moral intuitions about gambling are powerful. Even among people who accept that markets are efficient information-processing mechanisms, the association with "gambling" triggers deep-seated moral objections. Prediction markets exist in an uncomfortable space between finance (respectable) and gambling (suspect), and this ambiguity has plagued them throughout their history.

The controversy did not kill the idea. Despite the immediate political defeat, FutureMAP generated enormous publicity for prediction markets. Many people who had never heard of the IEM or information markets learned about them through the controversy. Within a few years, commercial prediction markets were thriving. The DARPA debacle may have slowed government adoption of prediction markets, but it accelerated private-sector interest.

Robin Hanson later reflected that the controversy was "the best and worst thing that ever happened to prediction markets" — it destroyed one promising project but put the concept on the map for a much larger audience.

2.4 The Rise of Play-Money Markets

2.4.1 Hollywood Stock Exchange

Even before the FutureMAP controversy, entrepreneurs had been experimenting with play-money prediction markets — platforms where participants trade virtual currency rather than real money. The most successful early example was the Hollywood Stock Exchange (HSX), launched in 1996 by Max Keiser and Michael Burns.

HSX allowed users to trade virtual shares in movies and movie stars. The price of a movie "stock" was tied to the film's actual box office performance, while "star bonds" reflected an actor's commercial track record. The platform attracted hundreds of thousands of users and generated remarkably accurate forecasts. Research by David Pennock (then at NEC Research Institute) and others showed that HSX prices predicted box office performance with accuracy comparable to professional analysts.

HSX was important for several reasons:

• It demonstrated that play-money markets could work. Skeptics had argued that prediction markets required real financial stakes to incentivize honest participation. HSX showed that reputation, competition, and entertainment value could substitute for real money — at least in some contexts.
• It attracted a mass audience. While the IEM had hundreds of participants, HSX had hundreds of thousands. This provided far richer data about market behavior and dynamics.
• It opened the door to broader applications. Because play-money markets avoided the legal complications of real-money gambling, they could address a much wider range of questions.

Cantor Fitzgerald (the financial services firm) acquired HSX in 2001, with plans to launch a real-money version for trading movie futures. The CFTC initially approved the concept but ultimately abandoned it in 2012 after objections from the Motion Picture Association of America, which feared that futures markets could be manipulated by studios.

2.4.2 NewsFutures and Other Early Platforms

Several other play-money platforms emerged in the late 1990s and early 2000s:

• NewsFutures (founded 1999 by Emile Servan-Schreiber): A play-money market focused on current events, particularly politics and technology. NewsFutures was notable for its focus on calibration and accuracy measurement.
• Foresight Exchange (founded 1994): One of the earliest online prediction markets, trading claims about science, technology, and current events. It operated entirely on play money and attracted a small but dedicated community of forecasters.
• TradeSports/BetFair: While not strictly play-money platforms (they operated as betting exchanges), these platforms pioneered the exchange model where participants bet against each other rather than against a bookmaker.

2.4.3 The Good Judgment Project Connection

The play-money tradition fed directly into one of the most important developments in forecasting science: the Good Judgment Project (GJP), founded by Philip Tetlock and Barbara Mellers in 2011.

Tetlock had spent decades studying the accuracy of expert predictions. His landmark 2005 book Expert Political Judgment: How Good Is It? How Can We Know? showed that experts were, on average, only slightly better than chance at predicting geopolitical events — and worse than simple statistical models. This finding was devastating to the credibility of pundit forecasting but opened the door to an obvious question: could prediction markets or structured forecasting methods do better?

The GJP was funded by the Intelligence Advanced Research Projects Activity (IARPA) — the intelligence community's research arm, which had taken up the baton dropped by DARPA after the FutureMAP debacle. IARPA organized a forecasting tournament in which teams competed to predict geopolitical events. The GJP recruited thousands of volunteer forecasters from the general public and tested various methods for aggregating their predictions.

The results were remarkable. The GJP's best forecasters — dubbed "superforecasters" — consistently outperformed intelligence analysts who had access to classified information. And the GJP's methods for aggregating predictions — which drew heavily on prediction market design — produced forecasts that were approximately 30% more accurate than those of the intelligence community.

While the GJP was not itself a prediction market (it used surveys and algorithms rather than trading), it validated many of the same principles. The connection between play-money prediction markets and the forecasting movement would continue to deepen in subsequent years, with platforms like Metaculus and Manifold Markets explicitly bridging the two traditions.

2.5 Real-Money Pioneers: InTrade and TradeSports

2.5.1 Founding and Early Growth

In 1999, two Irish entrepreneurs — John Delaney and Ron Bernstein — launched TradeSports, a sports betting exchange based in Dublin, Ireland. Operating under Irish law (which permitted licensed betting operations), TradeSports allowed users worldwide to trade contracts on sporting events using the exchange model, where participants bet against each other rather than against the house.

In 2001, Delaney launched InTrade as a sister platform focused on non-sports events: politics, economics, entertainment, and current affairs. InTrade quickly became the world's most prominent real-money prediction market. Its contract on the 2004 U.S. presidential election between George W. Bush and John Kerry became a media sensation, with major news networks routinely citing InTrade prices alongside poll numbers.

InTrade's appeal was simple: it offered a transparent, real-time window into collective expectations about important events. While polls measured what voters said they would do, InTrade measured what traders were willing to stake their money on. The distinction mattered — and the media loved the narrative of "Wall Street meets elections."

2.5.2 Key Moments

InTrade's history is marked by several notable episodes:

The 2004 U.S. Presidential Election. InTrade's election market attracted enormous attention and substantial trading volume. On election night, InTrade correctly signaled Bush's victory hours before the networks called the race, as returns from Ohio tipped the market decisively in Bush's favor.

The 2008 Financial Crisis. InTrade hosted contracts on the likelihood of a U.S. recession, the passage of the bank bailout (TARP), and the bankruptcy of major financial institutions. These contracts provided real-time public information that was otherwise available only through opaque credit default swap markets.

The 2008 and 2012 Presidential Elections. InTrade became a standard reference point for election forecasting, competing with Nate Silver's FiveThirtyEight model for media attention. In 2008, InTrade correctly predicted the winner of every state except Indiana and Missouri (which were extremely close). In 2012, InTrade's final prices gave Obama approximately a 67% chance of winning — accurate in direction, though somewhat less confident than Silver's 90% estimate.

The Saddam Hussein Capture Contract. In 2003, InTrade listed a contract on whether Saddam Hussein would be captured by a specific date. The price spiked dramatically on the afternoon of December 13, 2003 — hours before the official announcement of Hussein's capture. This was cited as evidence both of the market's remarkable information-processing speed and of the risk of insider trading on prediction markets.

2.5.3 Regulatory Challenges

InTrade's position was always legally precarious for U.S. participants. Under the Commodity Exchange Act, event contracts that do not involve a commodity with a deliverable supply can be classified as illegal off-exchange futures contracts. InTrade argued that it was regulated by Irish authorities and therefore beyond the reach of U.S. law, but this argument was never tested in court.

In 2005, the CFTC filed a civil complaint against TradeSports for offering commodity futures contracts to U.S. customers without proper registration. TradeSports settled the complaint by paying a fine and agreeing not to offer commodity-linked contracts to U.S. residents.

The Dodd-Frank Wall Street Reform and Consumer Protection Act, passed in 2010, gave the CFTC explicit authority to regulate event contracts and empowered the agency to block contracts that involved "activity unlawful under any Federal or State law," including contracts on terrorism, assassination, and war. This provision — directly inspired by the FutureMAP controversy — had implications far beyond the narrow categories it explicitly addressed.

In November 2012, the CFTC filed a civil complaint against InTrade, alleging that the company had offered illegal off-exchange options trading to U.S. customers and had failed to comply with the 2005 TradeSports settlement. InTrade was ordered to stop accepting U.S. customers.

2.5.4 Closure

On March 10, 2013, InTrade suspended all trading and announced that it was investigating "possible financial irregularities." The announcement was abrupt and alarming to traders, many of whom had substantial balances on the platform. John Delaney had died in May 2011 while climbing Mount Everest, and the company had been struggling under new management.

The subsequent investigation revealed significant problems. Customer funds had been commingled with operating funds, and some money appeared to have been used for unauthorized purposes. InTrade was placed into liquidation under Irish law, and customers ultimately recovered only a fraction of their deposits.

InTrade's closure was a watershed moment for prediction markets. The largest, most prominent real-money platform had failed — and it had failed not because its markets were inaccurate but because of financial mismanagement and regulatory pressure. The lessons were clear: prediction markets needed not just good contract design but also robust corporate governance, regulatory compliance, and customer protection.

We explore InTrade's story in greater detail in Case Study 1 at the end of this chapter.

2.6 The Prediction Market Renaissance (2014–2020)

2.6.1 The Post-InTrade Landscape

InTrade's closure left a void in the prediction market ecosystem. For a brief period, there was no major real-money prediction market accessible to U.S. and international participants. But the underlying demand for prediction markets had not disappeared — and several new platforms moved to fill the gap, approaching the challenge from different angles.

2.6.2 Augur: The Blockchain Approach

In 2014, a team led by Jack Peterson and Joey Krug launched a crowdfunding campaign for Augur, a decentralized prediction market built on the Ethereum blockchain. Augur raised over $5 million — one of the largest cryptocurrency crowdfunding campaigns at the time — and launched its mainnet in July 2018.

Augur represented a fundamentally different approach to prediction markets. Instead of a centralized company operating a platform (like InTrade), Augur was a protocol — a set of smart contracts running on the Ethereum blockchain that anyone could use to create and trade prediction market contracts. There was no company to shut down, no server to seize, and no single point of regulatory control.

The theoretical appeal was enormous. Augur promised prediction markets that were: - Censorship-resistant: No government could shut them down. - Trustless: Smart contracts automatically settled bets, eliminating the risk of operator fraud. - Global: Anyone with an Ethereum wallet could participate. - Permissionless: Anyone could create a market on any topic.

In practice, Augur's first version struggled. The user experience was poor (requiring users to interact directly with the Ethereum blockchain), transaction costs were high (Ethereum "gas" fees could exceed the value of small trades), and liquidity was thin (most markets had very few participants). Resolution disputes — determining who won a bet — proved surprisingly difficult to decentralize, leading to a complex and sometimes contentious oracle system.

Despite these difficulties, Augur was significant as a proof of concept. It demonstrated that decentralized prediction markets were technically feasible and laid the groundwork for later blockchain-based platforms, including Polymarket.

2.6.3 PredictIt's No-Action Letter

In 2014, Victoria University of Wellington in New Zealand applied to the CFTC for a no-action letter — the same regulatory instrument that had enabled the IEM. The university proposed to operate PredictIt, a real-money prediction market focused on political events, as an academic research project.

The CFTC granted the no-action letter in October 2014, with conditions similar to those governing the IEM: individual investments capped at $850 per contract, no more than 5,000 traders per market, and the requirement that the platform serve a bona fide academic research purpose.

PredictIt launched in 2015, operated by Aristotle International (a political data firm) under the Victoria University umbrella. It quickly became the dominant U.S. prediction market for political events, particularly during the tumultuous 2016 presidential election. PredictIt's markets on the Trump-Clinton race attracted enormous attention, with the platform's prices regularly cited in media coverage.

PredictIt was not without controversy. Critics argued that the academic research justification was thin and that the platform was, in practice, a commercial operation using the university affiliation as regulatory cover. The $850 investment limit and 5,000-trader cap also created liquidity constraints that sometimes led to distorted prices.

In August 2022, the CFTC withdrew PredictIt's no-action letter, ordering the platform to wind down its operations. PredictIt challenged the withdrawal in federal court, and as of early 2025, the legal battle continued. The case raised important questions about the CFTC's authority over prediction markets and the stability of the no-action letter framework.

2.6.4 Academic Developments

The period from 2014 to 2020 also saw significant academic progress in prediction market research:

• Market scoring rules: Robin Hanson's work on logarithmic market scoring rules (LMSR) — automated market makers that can provide liquidity even in thin markets — was refined and implemented in multiple platforms.
• Combinatorial prediction markets: Researchers developed methods for trading complex combinations of events (e.g., "What is the probability of a recession given that the incumbent party wins the election?"), enabling richer conditional forecasting.
• Behavioral research: Studies identified systematic biases in prediction markets, including the favorite-longshot bias, partisan bias, and the tendency for thin markets to be poorly calibrated.
• Comparison studies: Large-scale studies comparing prediction markets to polls, expert surveys, and statistical models provided increasingly nuanced evidence about when markets excel and when they falter.

2.7 The Modern Era: Polymarket, Kalshi, and the Mainstream (2020–Present)

2.7.1 The 2020 Turning Point

The 2020 U.S. presidential election, the COVID-19 pandemic, and the broader "information crisis" of the early 2020s created an environment in which prediction markets became more relevant than ever. Traditional media was widely distrusted, polls had been criticized for missing key dynamics in 2016, and social media was awash in misinformation. Prediction markets offered something that few other information sources could: continuously updated, incentive-driven probability estimates.

2.7.2 Polymarket's Rise

Polymarket, founded by Shayne Coplan in 2020, emerged as the dominant prediction market platform of the modern era. Built on the Polygon blockchain (a layer-2 scaling solution for Ethereum), Polymarket addressed many of the usability problems that had plagued Augur while retaining the advantages of blockchain settlement.

Key features of Polymarket's design:

• USDC settlement: Traders deposited and withdrew funds in USDC, a stablecoin pegged to the U.S. dollar, eliminating the volatility risk associated with cryptocurrency.
• Smooth user experience: Unlike Augur, Polymarket offered a web-based interface that was accessible to non-technical users.
• High liquidity: Polymarket's contracts on major events (particularly U.S. elections) attracted millions of dollars in trading volume.
• Conditional markets: Polymarket offered markets on conditional outcomes, enabling more nuanced forecasting.

Polymarket's growth was explosive. During the 2024 U.S. presidential election, Polymarket became a fixture of media coverage, with its probability estimates cited by major news outlets alongside polling averages. The platform's market on the presidential race attracted over $1 billion in cumulative trading volume — an unprecedented figure for a prediction market.

However, Polymarket also faced significant challenges:

• Regulatory ambiguity: In January 2022, the CFTC settled with Polymarket for operating an illegal unregistered trading platform, imposing a $1.4 million fine. Polymarket subsequently blocked U.S. users from its platform (though enforcement of this restriction was imperfect).
• Manipulation concerns: The 2024 election market saw episodes where large trades by individual participants (notably a trader known as "Fredi9999" or the "French whale") moved prices significantly, raising questions about whether prices reflected genuine collective judgment or the opinions of a few wealthy individuals.
• Thin markets: While Polymarket's highest-profile markets attracted substantial liquidity, many smaller markets had too few participants to produce reliable prices.

Despite these challenges, Polymarket demonstrated that prediction markets could scale to mainstream attention and that blockchain technology could provide a viable infrastructure for decentralized market operations.

2.7.3 Kalshi's CFTC Approval

While Polymarket operated in a regulatory gray zone, Kalshi took a different approach: full regulatory compliance. Founded in 2018 by Tarek Mansour and Luana Lopes Lara, Kalshi obtained designation as a Designated Contract Market (DCM) from the CFTC in November 2020 — the first prediction market platform to receive this status.

CFTC designation meant that Kalshi could legally offer event contracts to U.S. residents, subject to CFTC oversight. The company launched in July 2021 with markets on a range of topics, including weather events, economic indicators, and entertainment outcomes.

Kalshi's most significant regulatory battle involved political event contracts. In 2023, Kalshi applied to the CFTC for permission to list contracts on congressional elections. The CFTC initially rejected the application, arguing that political event contracts fell within the Dodd-Frank Act's prohibition on contracts involving "activity unlawful under any Federal or State law" (specifically, gambling on elections, which is prohibited in many states).

Kalshi challenged the rejection in federal court. In September 2024, a federal judge ruled in Kalshi's favor, finding that the CFTC had exceeded its authority in blocking the contracts. The CFTC's emergency stay was denied, and Kalshi began listing political event contracts. This was a landmark moment in the regulatory history of prediction markets, potentially opening the door to broad, legal, regulated prediction market trading in the United States.

2.7.4 Metaculus and the Forecasting Movement

Metaculus, founded in 2015 by Anthony Aguirre and Greg Laughlin, represents a different branch of the prediction market family tree. Rather than a trading platform, Metaculus is a forecasting community where participants submit probability estimates on questions about science, technology, politics, and global affairs. There is no money — real or play — at stake. Instead, participants are motivated by reputation scores, intellectual challenge, and community engagement.

Metaculus has become a focal point of the broader "forecasting movement" — a community of practice that draws on prediction market theory, superforecasting techniques, and effective altruism philosophy. The platform hosts thousands of questions, many focused on long-term risks (AI safety, climate change, pandemic preparedness) that traditional prediction markets rarely address.

Metaculus is significant because it demonstrates that the information-aggregation principles underlying prediction markets can work even without financial incentives, provided that other motivational structures (reputation, community, intellectual satisfaction) are sufficiently strong.

2.7.5 Manifold Markets: Play-Money Innovation

Manifold Markets, launched in 2022 by Stephen Grugett and James Grugett, brought fresh energy to the play-money prediction market concept. Manifold allows anyone to create a market on any question, using the platform's play-money currency ("mana"). The result is a vast, eclectic marketplace covering everything from geopolitics to personal bets between friends.

Manifold's key innovations include:

• Frictionless market creation: Any user can create a market in seconds, with no approval process. This has led to an explosion of niche markets that would never be viable on a real-money platform.
• Social features: Manifold integrates social media elements (comments, profiles, leaderboards) that make prediction trading a social activity.
• Subsidized markets: Market creators can add liquidity to encourage participation, effectively "sponsoring" questions they want the community to answer.
• API access: Manifold provides a robust API that has attracted a community of bot builders and algorithmic traders, further increasing liquidity.

As of 2025, Manifold had experimented briefly with real-money trading (through a partnership with a regulated entity) before reverting to play money, illustrating the persistent regulatory challenges facing prediction markets.

2.8 Academic Contributions and Key Researchers

2.8.1 Robin Hanson: The Theoretical Pioneer

No individual has contributed more to the theoretical foundations of prediction markets than Robin Hanson, an economist at George Mason University. Hanson's contributions span decades and cover nearly every aspect of prediction market theory:

• Idea futures (1990s): Hanson proposed using prediction markets to settle scientific and policy disputes, coining the term "idea futures."
• Market scoring rules (2003): Hanson developed the logarithmic market scoring rule (LMSR), which allows a single market maker to provide liquidity in prediction markets with arbitrarily many outcomes. The LMSR became the standard automated market maker for prediction markets and directly influenced the design of platforms like Augur and Manifold.
• Decision markets (2007): Hanson proposed "futarchy" — a governance system in which prediction markets are used to evaluate policy proposals. Under futarchy, elected officials would define national goals, and prediction markets would determine which policies are most likely to achieve those goals.
• Combinatorial markets: Hanson developed methods for trading complex conditional contracts, enabling prediction markets to address "if-then" questions.

2.8.2 Justin Wolfers and Eric Zitzewitz

Justin Wolfers (University of Michigan) and Eric Zitzewitz (Dartmouth College) co-authored some of the most influential empirical papers on prediction markets. Their 2004 paper "Prediction Markets" in the Journal of Economic Perspectives provided a comprehensive overview that became the standard introduction to the field. Key contributions include:

• Demonstrating that prediction market prices can be interpreted directly as probabilities under certain conditions.
• Documenting the accuracy of InTrade, the IEM, and other platforms across a wide range of events.
• Analyzing the potential for manipulation in prediction markets and finding that manipulation attempts generally fail because other traders quickly correct distorted prices.
• Proposing the use of prediction markets for corporate decision-making and government policy.

2.8.3 Kenneth Arrow and the Nobel Laureate Connection

Kenneth Arrow, the Nobel Prize-winning economist, lent his enormous prestige to the prediction market cause. In 2008, Arrow co-authored a letter to Science (with Robert Forsythe, Michael Gorham, Robert Hahn, Robin Hanson, John Ledyard, Saul Levmore, Robert Litan, Paul Milgrom, Forrest Nelson, George Neumann, Marco Ottaviani, Thomas Rietz, Thomas Schelling, Robert Shiller, Vernon Smith, Erik Snowberg, Cass Sunstein, Paul Tetlock, Philip Tetlock, Hal Varian, Justin Wolfers, and Eric Zitzewitz) calling for the legal barriers to prediction markets to be lowered. The letter, titled "The Promise of Prediction Markets," argued that prediction markets were a valuable social institution being held back by outdated gambling regulations.

The involvement of Arrow — along with other Nobel laureates such as Vernon Smith (2002 Nobel, for experimental economics) and Robert Shiller (2013 Nobel, for empirical analysis of asset prices) — gave prediction markets a degree of academic legitimacy that helped counter the "gambling" stigma.

2.8.4 Charles Manski and the Skeptical View

Not all prominent economists embraced prediction markets uncritically. Charles Manski (Northwestern University) published an important critique in 2006, arguing that prediction market prices cannot, in general, be interpreted as probabilities. Manski showed that the relationship between market prices and true probabilities depends on the risk preferences of market participants — a point that is technically correct but whose practical significance is debated.

Manski's critique was important because it pushed prediction market researchers to be more rigorous about the conditions under which market prices are well-calibrated. It also highlighted the limitations of prediction markets in thin or highly risk-averse environments.

2.8.5 Key Papers Timeline

2.9 Lessons from History

2.9.1 What Failed Markets Teach Us

The history of prediction markets is littered with failures — platforms that launched with great promise and then collapsed, stagnated, or were shut down. These failures are as instructive as the successes:

InTrade's failure was about governance, not markets. InTrade's markets worked well. Its forecasts were accurate, its user base was engaged, and its contracts were widely cited. The platform failed because of financial mismanagement and regulatory defiance. The lesson: technical excellence in market design is necessary but not sufficient. Prediction market platforms also need trustworthy operators, transparent finances, and viable regulatory strategies.

Augur v1's failure was about user experience. Augur's underlying technology was sound — decentralized, censorship-resistant, and transparent. But the user experience was terrible. High transaction costs, slow settlement times, and a confusing interface drove away all but the most dedicated users. The lesson: prediction markets must be easy to use. Blockchain technology is valuable only if it serves users, not the other way around.

FutureMAP's failure was about communication. The Policy Analysis Market was a reasonable application of prediction market theory to intelligence analysis. It was killed by a communications failure — the inability to explain the project's purpose before opponents framed it as a "terrorism betting market." The lesson: prediction markets are politically and emotionally sensitive. Advocates must invest in clear, proactive communication.

2.9.2 Regulatory Patterns

A clear pattern emerges from the regulatory history of prediction markets:

1. Innovation outpaces regulation. New prediction market platforms typically launch before regulators have developed clear rules. This creates a period of legal ambiguity that can last years or decades.

2. Regulation is reactive, not proactive. Regulators generally do not act until a prediction market becomes prominent enough to attract political attention. The CFTC did not move against InTrade until the platform had been operating for over a decade.

3. Academic exemptions create wedge openings. The IEM and PredictIt both obtained regulatory permission by emphasizing their academic research purposes. These exemptions created precedents that later commercial platforms could build on.

4. The CFTC is the key gatekeeper. In the United States, the Commodity Futures Trading Commission has emerged as the primary regulator of prediction markets. The CFTC's decisions about which event contracts to permit have been the most consequential regulatory actions in the field.

5. Blockchain creates new regulatory challenges. Decentralized prediction markets like Augur and Polymarket operate outside the traditional regulatory framework, forcing regulators to develop new approaches.

2.9.3 Technology Evolution

The technology underlying prediction markets has evolved dramatically:

Each technological generation expanded what was possible. The web enabled global participation. Blockchain enabled decentralization and censorship resistance. Layer-2 scaling solutions and stablecoins enabled low-cost, dollar-denominated trading. The next generation — potentially involving AI-powered market making and cross-chain interoperability — promises further expansion.

2.9.4 The Recurring Tension: Accuracy vs. Legitimacy

Throughout their history, prediction markets have been caught in a recurring tension between their demonstrated accuracy and their contested legitimacy. The evidence that markets produce good forecasts is strong and growing. But public acceptance has been slow and uneven, hampered by associations with gambling, moral objections to profiting from negative events, and fears of manipulation.

This tension has shaped the development of the field in fundamental ways. It has pushed some platforms toward play money (avoiding gambling regulations but potentially sacrificing accuracy). It has driven others toward full regulatory compliance (gaining legitimacy but accepting constraints on market design). And it has motivated researchers to develop alternatives — like Metaculus's reputation-based system — that capture the information-aggregation benefits of markets without the financial elements that trigger moral objections.

Understanding this tension is essential for anyone who wants to work with prediction markets. Technical skill in market design is necessary but not sufficient. Success also requires navigating the political, regulatory, and cultural landscape in which prediction markets operate.

2.10 Building a Historical Timeline in Python

Now let us put our historical knowledge to practical use. In this section, we will build an interactive timeline visualization of prediction market history using Python and matplotlib. This exercise serves two purposes: it reinforces the historical content of this chapter, and it introduces basic data visualization techniques that we will build on in later chapters.

2.10.1 Setting Up the Data

First, we need to structure our historical data. We will create a pandas DataFrame with key events, categorized by type:

import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from datetime import datetime
import numpy as np

# Define prediction market history events
events = [
    # Ancient and Pre-Modern
    {"year": 1503, "event": "Documented papal election betting in Rome",
     "category": "Ancient/Pre-Modern", "importance": 2},
    {"year": 1688, "event": "Lloyd's Coffee House opens (London)",
     "category": "Ancient/Pre-Modern", "importance": 2},
    {"year": 1750, "event": "Jockey Club founded; organized betting matures",
     "category": "Ancient/Pre-Modern", "importance": 1},
    {"year": 1868, "event": "U.S. political betting markets flourish (NYC)",
     "category": "Ancient/Pre-Modern", "importance": 2},

    # Academic Milestones
    {"year": 1945, "event": "Hayek publishes 'The Use of Knowledge in Society'",
     "category": "Academic", "importance": 3},
    {"year": 1988, "event": "Iowa Electronic Markets (IEM) founded",
     "category": "Academic", "importance": 3},
    {"year": 1990, "event": "Hanson proposes 'idea futures'",
     "category": "Academic", "importance": 2},
    {"year": 2003, "event": "Hanson develops LMSR market scoring rule",
     "category": "Academic", "importance": 2},
    {"year": 2004, "event": "Wolfers & Zitzewitz publish 'Prediction Markets'",
     "category": "Academic", "importance": 3},
    {"year": 2005, "event": "Tetlock publishes Expert Political Judgment",
     "category": "Academic", "importance": 3},
    {"year": 2008, "event": "Arrow et al. 'Promise of Prediction Markets'",
     "category": "Academic", "importance": 2},
    {"year": 2011, "event": "Good Judgment Project begins",
     "category": "Academic", "importance": 3},
    {"year": 2015, "event": "Tetlock publishes Superforecasting",
     "category": "Academic", "importance": 2},

    # Commercial Platforms
    {"year": 1996, "event": "Hollywood Stock Exchange launches",
     "category": "Commercial", "importance": 2},
    {"year": 1999, "event": "TradeSports founded (Dublin)",
     "category": "Commercial", "importance": 2},
    {"year": 2001, "event": "InTrade launches",
     "category": "Commercial", "importance": 3},
    {"year": 2014, "event": "PredictIt receives CFTC no-action letter",
     "category": "Commercial", "importance": 3},
    {"year": 2015, "event": "Metaculus founded",
     "category": "Commercial", "importance": 2},
    {"year": 2018, "event": "Augur v1 launches on Ethereum",
     "category": "Commercial", "importance": 2},
    {"year": 2020, "event": "Polymarket founded",
     "category": "Commercial", "importance": 3},
    {"year": 2020, "event": "Kalshi receives CFTC DCM designation",
     "category": "Commercial", "importance": 3},
    {"year": 2022, "event": "Manifold Markets launches",
     "category": "Commercial", "importance": 2},

    # Regulatory Events
    {"year": 2003, "event": "DARPA FutureMAP canceled",
     "category": "Regulatory", "importance": 3},
    {"year": 2005, "event": "CFTC action against TradeSports",
     "category": "Regulatory", "importance": 2},
    {"year": 2010, "event": "Dodd-Frank Act passes (event contract provisions)",
     "category": "Regulatory", "importance": 2},
    {"year": 2012, "event": "CFTC sues InTrade",
     "category": "Regulatory", "importance": 2},
    {"year": 2013, "event": "InTrade closes",
     "category": "Regulatory", "importance": 3},
    {"year": 2022, "event": "CFTC settles with Polymarket ($1.4M fine)",
     "category": "Regulatory", "importance": 2},
    {"year": 2022, "event": "CFTC withdraws PredictIt no-action letter",
     "category": "Regulatory", "importance": 2},
    {"year": 2024, "event": "Court rules Kalshi can list election contracts",
     "category": "Regulatory", "importance": 3},
]

df = pd.DataFrame(events)
print(f"Total events in timeline: {len(df)}")
print(f"Categories: {df['category'].unique()}")

2.10.2 Creating the Timeline Visualization

Now we build the visualization. We use a vertical timeline layout with color-coded categories and size-coded importance levels:

def create_prediction_market_timeline(df, save_path=None):
    """
    Create a visually rich timeline of prediction market history.

    Parameters
    ----------
    df : pd.DataFrame
        DataFrame with columns: year, event, category, importance
    save_path : str, optional
        File path to save the figure
    """
    # Color scheme for categories
    colors = {
        "Ancient/Pre-Modern": "#8B4513",
        "Academic": "#1E90FF",
        "Commercial": "#228B22",
        "Regulatory": "#DC143C",
    }

    fig, ax = plt.subplots(figsize=(14, 20))

    # Draw the central timeline spine
    ax.axvline(x=0.5, color="gray", linewidth=2, alpha=0.3)

    # Sort events by year
    df_sorted = df.sort_values("year").reset_index(drop=True)

    # Alternate events left and right
    for i, row in df_sorted.iterrows():
        side = -1 if i % 2 == 0 else 1
        x_text = 0.5 + side * 0.35
        ha = "right" if side == -1 else "left"

        # Size the marker by importance
        marker_size = row["importance"] * 40

        # Plot the marker on the spine
        y_pos = i
        ax.scatter(0.5, y_pos, s=marker_size,
                   color=colors[row["category"]],
                   zorder=5, edgecolors="white", linewidth=0.5)

        # Draw connector line
        ax.plot([0.5, x_text], [y_pos, y_pos],
                color=colors[row["category"]], linewidth=0.8, alpha=0.5)

        # Add event text
        label = f"{row['year']}: {row['event']}"
        ax.text(x_text, y_pos, label, ha=ha, va="center",
                fontsize=8, color=colors[row["category"]],
                fontweight="bold" if row["importance"] == 3 else "normal")

    # Add legend
    for cat, color in colors.items():
        ax.scatter([], [], c=color, s=80, label=cat)
    ax.legend(loc="lower right", fontsize=10, framealpha=0.9)

    # Formatting
    ax.set_xlim(-0.1, 1.1)
    ax.set_ylim(-1, len(df_sorted))
    ax.axis("off")
    ax.set_title("A History of Prediction Markets",
                 fontsize=18, fontweight="bold", pad=20)

    plt.tight_layout()
    if save_path:
        plt.savefig(save_path, dpi=150, bbox_inches="tight")
        print(f"Timeline saved to {save_path}")
    plt.show()


# Generate the timeline
create_prediction_market_timeline(df, save_path="prediction_market_timeline.png")

2.10.3 Adding Interactivity with Annotations

For a more detailed exploration, we can create an annotated version with tooltips. While true interactivity requires JavaScript or a dedicated library, we can simulate it with matplotlib's annotation capabilities:

def create_era_summary(df):
    """
    Summarize events by era and category.
    """
    # Define eras
    era_bins = [0, 1900, 1995, 2005, 2014, 2020, 2030]
    era_labels = [
        "Pre-Modern (before 1900)",
        "Early Digital (1988-1995)",
        "Growth & Controversy (1996-2005)",
        "Maturation (2006-2014)",
        "Renaissance (2015-2020)",
        "Modern Era (2021-present)",
    ]

    df["era"] = pd.cut(df["year"], bins=era_bins, labels=era_labels)

    summary = df.groupby(["era", "category"]).size().unstack(fill_value=0)
    print("\nEvents by Era and Category:")
    print(summary)
    print(f"\nTotal events: {len(df)}")

    return summary


summary = create_era_summary(df)

The complete code for this visualization is available in code/example-01-timeline-visualization.py. In Chapter 3, we will build on these techniques to create live, data-driven visualizations of active prediction markets.

2.11 Chapter Summary

This chapter has traced the history of prediction markets from their ancient origins to the modern era. Here are the key themes:

1. The idea is old; the technology is new. People have been using financial incentives to aggregate information about future events for centuries. What has changed is the technology — from coffee-house wagers to electronic exchanges to blockchain protocols — and the scale of participation.

2. Accuracy is well-documented. The Iowa Electronic Markets, InTrade, Polymarket, and other platforms have compiled track records that demonstrate prediction markets can forecast events accurately, often outperforming polls and expert judgment.

3. Regulation is the central challenge. Every major prediction market has faced regulatory obstacles. The regulatory environment has been shaped by the tension between the demonstrated value of prediction markets and the political, moral, and legal objections to what is often framed as "gambling."

4. Multiple models coexist. Real-money markets (Kalshi), cryptocurrency markets (Polymarket), play-money markets (Manifold), and reputation-based systems (Metaculus) each have strengths and weaknesses. The field has not converged on a single model.

5. Academic research provides the foundation. The work of Hanson, Wolfers, Zitzewitz, Arrow, Tetlock, and others has given prediction markets a theoretical and empirical foundation that distinguishes them from mere gambling.

6. Failure teaches as much as success. InTrade's collapse, Augur's usability problems, and the FutureMAP controversy all contain lessons about what prediction markets need to succeed: not just good market design, but also good governance, good communication, and viable regulatory strategies.

Key Terms Revisited

What's Next

In Chapter 3: How Prediction Markets Work, we will move from history to mechanics. You will learn exactly how prediction market contracts are structured, how prices are formed through order books and automated market makers, and how to interpret market probabilities. We will build a simple order-book matching engine in Python and use it to simulate market dynamics. The historical context from this chapter will help you understand why modern markets are designed the way they are — and what trade-offs their designers faced.

Chapter 2 is part of Learning Prediction Markets — From Concepts to Strategies. Previous: Chapter 1 — What Are Prediction Markets? Next: Chapter 3 — How Prediction Markets Work