a16z Crypto's latest article: Why do we need prediction markets?

Author: Scott Kominers

Compiled by: Jiahua, ChainCatcher

Prediction markets allow people to trade on the outcomes of events. Last year, they entered the U.S. on a large scale and are now used to track various events from geopolitics to entertainment award results. But what exactly are they?

As an economist who has long studied markets and incentive mechanisms, my answer is simple: prediction markets are essentially markets. Markets are the fundamental tool for allocating resources, ensuring that goods and services flow to those who value them the most.

In this process, markets also aggregate information: market clearing (i.e., reaching supply-demand equilibrium) is essentially a mechanism for summarizing the perceptions of all participants and refining them into price signals.

Prediction market platforms and products directly utilize this information aggregation capability to predict specific future events: they design an asset linked to an event that generates returns once a specific outcome occurs, and people trade this asset based on their judgment of whether the outcome will happen.

Such usage has existed for a long time.

Companies have long used prediction markets to obtain tacit information from employees, such as predicting whether an important product will be released on time; scientists use them to assess which experiments are likely to be successfully replicated; and now several media outlets collaborate with prediction markets to supplement their information sources and reporters' reports with "crowd wisdom."

Prediction markets collect information directly from participants, which is everyone's judgment about the future, and aggregate this information into a market to answer how likely a certain event is to occur.

People can "bet" on the future value of a company in the stock market or "bet" on the future price of commodities like oil in the same way. The difference is that the demand for assets like oil is influenced by many factors simultaneously, while the assets designed by prediction markets only generate returns when a specific event occurs.

If oil prices rise, we know that demand has increased relative to supply, but we may not know the underlying reasons: it could be that people expect an escalation in Middle Eastern conflicts, or it could be that someone has found a new use for oil.

With prediction markets, you can isolate each possibility for prediction.

For example, a prediction market regarding "whether the Strait of Hormuz will remain open at a specific point in time" could revolve around a contract: once the event occurs, each unit of the contract pays one dollar.

As people repeatedly buy and sell this asset, the market price becomes a "probability indicator," reflecting traders' overall judgment of the likelihood of the event occurring.

How does it work specifically? Suppose the unit market price for a certain outcome is $0.50, equivalent to a 50% probability. If you believe the likelihood of the strait remaining open is higher than 50%, say 67%, you would buy in; once your judgment is correct, you would obtain a total return of $0.67 at a cost of $0.50.

This purchase would also push up the market price and the corresponding probability estimate, effectively saying "someone thinks the market has underestimated it." Conversely, when someone believes the price is too high, they will sell at a lower price (or short), thus pulling down the overall probability estimate of the market.

When prediction markets operate well, they have several clear advantages over other prediction methods.

First, they can directly provide a probability estimate, and this alone is a "superpower."

Polls and surveys only provide "opinion proportions"; to convert this into probabilities, you need to perform statistical reasoning to determine the relationship between the measured proportion and the overall population. Moreover, polls are often just snapshots at a certain point in time, while prediction markets can update in real-time with the addition of new participants and new information.

More critically, prediction markets come with built-in incentives: both buyers and sellers have real money at stake, and a wrong bet results in a loss. This encourages participants to carefully weigh the information they have and invest in the issues they are most confident about.

Conversely, being able to profit from information and professional judgment in prediction markets will also motivate people to actively conduct research and clarify issues.

(A well-known example is that before the 2024 U.S. presidential election, a prediction market participant even conducted their own poll to uncover information that standard polling organizations could not obtain using an unconventional method.)

Finally, prediction markets also have a huge advantage in coverage. Someone who understands which events might affect oil demand can, in principle, go long or short on oil; however, many outcomes we want to predict do not have corresponding commodity or stock markets to bet on. In this case, prediction markets become an ideal choice.

For instance, recently a batch of prediction markets has emerged specifically to aggregate judgments on "which AI model performs best on various tasks." Such questions are too niche to be reflected in traditional commodity markets. Anyone can create and fund a prediction market for these kinds of specialized questions.

These ideas are not new. Similar practices existed as early as the 16th century in Europe, where people used them to predict the next pope.

The foundation of modern prediction markets lies in economics, statistics, market design, and computer science. Charles Plott and Shyam Sunder proposed the earliest formal academic framework in the 1980s, shortly after which the first modern prediction market, the Iowa Electronic Markets, was born.

With the help of the internet, this model can now aggregate dispersed information from around the world. However, for prediction markets to truly realize their potential, several prerequisites must be met.

One category is infrastructure issues: how to verify and reach consensus on "whether a certain event has occurred," how to ensure market operations are transparent and auditable, and how to handle the settlement of contracts that may provoke disputes or even be manipulated on a large scale.

Another category involves challenges in market design. First, those who truly possess relevant information must be willing to participate. If participants are uninformed, price signals do not convey any meaningful information; conversely, only by involving those who hold various types of information can the estimates from prediction markets remain accurate.

I pointed out in 2016 that prediction markets may have underestimated the probabilities of Brexit and Trump's first election because the participants at that time did not sufficiently understand the rise of populism.

Another issue is that if someone possesses "perfect" information, such as knowing the real outcome in advance, this is equally problematic, especially if they can influence the outcome of the event.

Imagine this: if an insider from a secret meeting about the papal election goes to bet on the "next pope" prediction market, trading before the news of Leo's election is officially announced, or even tries to sway the election to ensure their chosen candidate wins, what would happen?

Because of this, once potential participants anticipate that insiders will be trading, the rational choice is to simply stay away, leading to the market's collapse.

Finally, some may deliberately distort the prices in prediction markets to influence public perception of the probability of a certain outcome, turning it from a tool for "aggregating beliefs" into a tool for "manipulating beliefs."

For example, if a candidate's PR team wants to convince the public that they are certain to win, they might use part of their campaign funds to sway the relevant markets.

However, in this regard, prediction markets have a certain self-correcting ability: once the probability of a contract is pushed to an unreasonable height, there will always be someone willing to take the other side of the trade.

All of this indicates that prediction markets need to achieve higher transparency and clarity in participant management, contract design, and operational levels. But as long as designers can solve these challenges, prediction markets are likely to become one of our core tools for forecasting the future.