On February 8th US time (7:30 AM Beijing Time on February 9th), hundreds of millions of NFL fans gathered in front of their screens to watch the Super Bowl, with many also keeping an eye on another screen—closely monitoring the trading dynamics of prediction markets, where betting categories encompass everything from championship outcomes and final scores to the passing yards of each team's quarterback.
Over the past year, the trading volume of US prediction markets reached at least $27.9 billion, covering a vast array of subjects, from sports event results and economic policy decisions to new product launches. However, the nature of these markets has always been controversial: Are they a form of trading or gambling? A tool for aggregating collective wisdom for news, or a means of scientific validation? And is the current development model already the optimal solution?
As an economist who has long studied markets and incentive mechanisms, my answer begins with a simple premise: prediction markets are, in essence, markets. And markets are core tools for allocating resources and integrating information. The operating logic of prediction markets is to launch assets linked to specific events—when the event occurs, traders holding the asset receive a payout. People then trade based on their own judgment of the event's outcome, thereby unleashing the core value of the market.
From a market design perspective, referring to information from prediction markets is far more valuable than trusting the opinion of a single sports commentator, or even looking at the betting odds from Las Vegas. The primary goal of traditional sports betting institutions is not to predict the outcome of games, but to 'balance the betting funds' by adjusting odds, attracting money to the side with less betting volume at any given moment. Las Vegas betting seeks to entice players to bet on underdog outcomes, whereas prediction markets enable people to execute trades based on their genuine judgment.
Prediction markets also make it easier to extract effective signals from vast amounts of information. For example, if you want to gauge the likelihood of new tariffs being imposed, deriving this from soybean futures prices would be an indirect process—as futures prices are influenced by multiple factors. But if you ask this question directly in a prediction market, you can get a more straightforward answer.
The prototype of this model can be traced back to 16th-century Europe, where people even placed bets on 'the next Pope.' The development of modern prediction markets is rooted in contemporary theories of economics, statistics, mechanism design, and computer science. In the 1980s, Charles Plott of Caltech and Shyam Sunder of Yale University established its formal academic framework, and soon after, the first modern prediction market—the Iowa Electronic Markets—was launched.
The mechanism of prediction markets is actually quite simple. Take the bet 'Will Seattle Seahawks quarterback Sam Darnold pass the ball within the opponent's one-yard line?' as an example. The market issues corresponding trading contracts; if the event occurs, each contract pays the holder $1. As traders continuously buy and sell this contract, the market price of the contract can be interpreted as the probability of the event occurring, representing the collective judgment of the traders. For instance, a contract priced at $0.50 implies the market believes there is a 50% chance the event will happen.
If you judge the probability of the event to be higher than 50% (say, 67%), you can buy this contract. If the event ultimately occurs, the contract you purchased for $0.50 yields a $1 payout, resulting in a gross profit of $0.67. Your buying action will push up the market price of the contract, and the corresponding probability estimate will also rise, sending a signal to the market: someone believes the current market underestimates the likelihood of the event. Conversely, if someone believes the market overestimates the probability, selling will drive down the price and the probability estimate.
When prediction markets function well, they demonstrate significant advantages over other forecasting methods. Opinion polls and surveys can only yield the proportion of views; converting these into probability estimates requires statistical methods to analyze the relationship between the survey sample and the overall population. Moreover, such survey results are often static data at a specific moment, whereas information in prediction markets continuously updates with the arrival of new participants and new information.
More crucially, prediction markets have clear incentive mechanisms; traders are truly 'skin in the game.' They must carefully sift through the information they possess and only invest funds and take risks in areas they understand best. In prediction markets, people can convert their information and expertise into profits, which also incentivizes them to proactively delve deeper into relevant information.
Finally, the coverage scope of prediction markets far surpasses that of other tools. For instance, someone with information affecting oil demand can profit by going long or short on crude oil futures. But in reality, many outcomes we wish to predict cannot be realized through commodity or stock markets. For example, specialized prediction markets have recently emerged attempting to aggregate various judgments to predict the solution time for specific mathematical problems—information crucial for scientific development and an important benchmark for measuring the progress of artificial intelligence.
Despite their significant advantages, prediction markets still need to resolve many issues to truly realize their value. First, at the market infrastructure level, there are persistent questions that need clarification: How to verify whether a specific event has truly occurred and achieve market consensus? How to ensure the transparency and auditability of market operations?
Next are the challenges in market design. For instance, there must be participants with relevant information entering to trade—if all participants are uninformed, the market price cannot convey any effective signal. Conversely, various participants holding different relevant information need to be willing to trade; otherwise, the valuation in prediction markets will be biased. The prediction market before the Brexit referendum is a typical counterexample.
Furthermore, if participants with absolute insider information enter the market, new problems arise. For example, the Seahawks' offensive coordinator knows exactly whether Sam Darnold will pass within the one-yard line and can even directly influence this outcome. If such individuals participate in trading, market fairness would be severely compromised. If potential participants believe there are insider traders in the market, they might rationally choose to stay away, ultimately leading to a market collapse.
Additionally, prediction markets also face the risk of manipulation: someone might turn this tool, originally intended for aggregating collective judgment, into a means of manipulating public opinion. For instance, a candidate's campaign team might use campaign funds to influence the valuation in prediction markets to create an atmosphere of 'impending victory.' Fortunately, prediction markets have some self-correcting ability in this regard—if the probability estimate of a contract deviates from a reasonable range, there will always be traders choosing to take the opposite position, bringing the market back to rationality.
Given the various risks mentioned above, prediction market platforms must strive to enhance operational transparency and clearly disclose the rules governing participant management, contract design, market operation, and other aspects. If these issues can be successfully resolved, we can foresee that prediction markets will play an increasingly important role in the future of forecasting.
