The prediction market is essentially a market.

Written by: Scott Duke Kominers, a16z crypto research partner

Translated by: Chopper, Foresight News

Prediction markets allow users to trade based on the outcomes of various events. These platforms began to be widely implemented in the United States last year, and today they cover a wide range of events from geopolitical issues to entertainment award winners. But what exactly is a prediction market?

As an economist who has long studied market mechanisms and incentive systems, my answer is simple: a prediction market is essentially an ordinary market. Markets are foundational tools for allocating resources, allowing goods and services to flow to those who need them the most. In this process, markets also possess the ability to aggregate information: the process of clearing supply and demand combines all the information held by participants and transforms it into signals like prices.

Prediction market platforms and related products directly utilize this information aggregation ability to anticipate the direction of specific future events. The platforms will introduce underlying assets corresponding to specific events, and as long as the preset outcome comes true, holders can earn returns. Users trade these assets based on their judgment of the probability of an event occurring. For a long time, many companies have relied on prediction markets to extract tacit information held by employees to assess whether key products can launch on time. Researchers also use this tool to evaluate which experimental conclusions are replicable. Nowadays, many media organizations are choosing to collaborate with prediction markets, using collective intelligence to supplement frontline reporting and traditional coverage, thus enriching content dimensions.

Prediction markets gather all participants' individual judgments about the future and integrate these viewpoints to form a trading market, thereby estimating the occurrence probabilities of various events. Users place bets on event outcomes in such markets, which is logically no different from predicting the stock prices of publicly listed companies in the stock market or trading oil prices in the commodities market. The difference is that the prices of assets like oil are influenced by multiple complex factors, whereas the underlying assets in prediction markets only yield returns when designated events occur.

When oil prices rise, we can deduce that current demand is greater than supply, but we may not know the reasons behind it: is the market concerned about escalating tensions in the Middle East or is there a new application scenario for oil? Prediction markets can set up trading underlyings targeting a single possibility, allowing for accurate disaggregation of predictions. For example, if a market is established for "Will the Strait of Hormuz be open for navigation at a specified time?" the corresponding contract rules could be set as: if the event happens, each contract pays out $1. As users continually buy and sell, the market price becomes a probability indicator, reflecting the collective judgment of all traders on the likelihood of the event occurring.

The logic of its operation is as follows: Assume the current price of each underlying is $0.50, representing the market's belief that the probability of the event occurring is fifty-fifty. If you believe the probability of navigation is above 50%, say reaching 67%, you can buy that underlying. Once judged correctly, the underlying you purchased at $0.50 can eventually yield $0.67 in return. This buying action would further push up the market price and estimated probability, indicating that there are traders who believe the market previously underestimated the likelihood of the event occurring. Conversely, if someone thinks the current price is too high, they may choose to sell low or short the underlying, thus driving down the market's probability valuation.

Compared to other prediction methods, well-functioning prediction markets have substantial advantages. First, they can directly output quantifiable probability results, which is a major highlight. Polls and surveys can only estimate the proportion of opinions, and any attempt to infer event probabilities from these requires combining statistical methods to analyze the relationship between sample data and the overall population. Moreover, poll results are mostly just static data at a certain time point, while prediction markets continuously update judgments in real time as new participants enter and new information emerges.

More importantly, prediction markets come with intrinsic incentive constraints. Both buyers and sellers invest real money, and if they make incorrect judgments, they incur losses. This compels participants to thoroughly review the information they possess, prioritizing participation in trades within areas they are familiar with and where they have informational advantages. Conversely, the desire to profit through information and expertise also prompts people to actively conduct research and dig deep into event-related clues. A well-known case is that, on the eve of the 2024 U.S. election, participants in prediction markets used unconventional methods to conduct polls in an attempt to gain information that traditional polling institutions could not access.

Finally, prediction markets have an incredibly broad scope. Theoretically, traders who possess information about the oil industry can express their judgments by going long or short on crude oil contracts, but in reality, there are numerous event outcomes that cannot be predicted through mainstream commodity markets or stock markets, making such scenarios perfectly suited for prediction markets. For example, many prediction markets have recently begun to launch relevant underlyings to comprehensively assess the performance of various artificial intelligence models in different tasks. Trends in such niche areas are hard to reflect in traditional commodity markets. Anyone can build and fund a prediction market to address these niche questions.

Prediction markets are not a new phenomenon; their early forms can be traced back to 16th-century Europe, where they were used to predict the selection of the next pope. Modern prediction markets integrate knowledge from multiple fields, including economics, statistics, market design, and computer science. In the 1980s, Charles Plott and Shyam Sundar first established a formal academic framework for this mechanism. Shortly thereafter, the world's first modern prediction market—the Iowa Electronic Market—was officially launched. Leveraging internet technology, this model has been able to integrate scattered information from around the world, continually growing in scale.

However, to fully unleash the potential of prediction markets, there are still many challenges that need to be addressed. First is the infrastructure aspect: how to determine the final result of an event and reach a consensus, how to ensure market operations are transparent and transactions are traceable; and what large-scale adjudication mechanisms to adopt when contract payout results are disputed or even subjected to manipulation.

Secondly, there are challenges on the market design front. First, the group holding core information must be involved. If all participants know nothing, then market price signals are of no reference value. Conversely, if informed parties are unwilling to participate, prediction outcomes will be biased. As I pointed out back in 2016, during the year of Brexit and Donald Trump's first election as President of the United States, prediction markets underestimated the likelihood of these events occurring because the participants at the time failed to recognize the trend of populism's rise.

Additionally, if individuals with insider information enter the market to trade, it can create risks, especially if these individuals also have the power to influence the outcome of the event. Imagine if internal members of the papal election meeting placed bets in the "next pope" prediction market beforehand, exploiting insider information to front-run trades; or even secretly intervening in the election outcome for their positions, the consequences are dire. If participants generally believe there is insider trading in the market, everyone will choose to exit, ultimately leading to the collapse of the entire market.

There's another risk: someone may deliberately manipulate prediction market prices to guide public perception of event probabilities. In such cases, prediction markets would transform from tools aggregating opinions to instruments for manipulating public opinion. For instance, a campaign team could use campaign funds to deliberately inflate the market probability of their own victory, creating an illusion of being ahead. However, prediction markets possess a certain degree of self-correcting ability: as long as prices deviate significantly from reasonable ranges, traders will make counter bets to hedge against unreasonable pricing.

All these issues indicate that prediction markets need further refinement of rules, clarifying standards for participant entry, contract design, and overall operation. However, if industry practitioners can solve these challenges one by one, prediction markets will ultimately become an important tool for humanity to predict the future and respond to uncertainties.