THE SUPER BOWL OF PREDICTION MARKETS / PROJECT SYNDICATE

The Super Bowl of Prediction Markets

As prediction markets have exploded onto the scene in recent years, allowing bets to be placed on a wide range of event-driven outcomes, many have asked whether it is optimal to have markets in everything. But the real question is how best to ensure that these powerful tools are transparent and auditable.

Scott Duke Kominers

CAMBRIDGE – This weekend, hundreds of millions of NFL football fans will settle in to watch the Super Bowl, and many will have their eyes on a second screen as well. 

They will be watching the trading on prediction markets, where bets will cover everything from the overall winner and the final score to the number of yards passed by each team’s quarterback.

Over the past year, prediction markets in the United States have generated at least $27.9 billion in trading volume, tracking everything from sporting events and economic policy decisions to new product launches. 

But their true nature has been debated. 

Are they about trading? 

Gambling? 

Do they serve as crowdsourced journalism or scientific replication tools? 

And is it optimal to have “markets in everything”?

As an economist who has long studied marketplaces and incentive mechanisms, my own answer starts with a simple premise: Prediction markets are markets, and markets are a fundamental tool for allocating resources and aggregating information. 

They harness this power by introducing an event-specific asset that pays off if a given outcome occurs, allowing people to place trades based on their own beliefs about what will happen.

From a market-design perspective, this is better than just following a single sports pundit, or even the Vegas line. 

A traditional sportsbook is not trying to predict who will win, but rather to “balance the action” by adjusting the odds to attract bets on whichever side has less money behind it at a given moment. 

Vegas wants to make people willing to bet on long shots, whereas a prediction market enables people to trade based on what they truly believe.

Prediction markets also make it easier to extract signals from the noise. 

If you want to estimate the likelihood of new tariffs, you could try to infer beliefs about the answer indirectly from the price of soybean futures, whose value reflects many forces at once. 

But it’s much more direct to pose the question through a prediction market.

Variations on this idea go back at least to 16th-century Europe, where bets were placed on who would be the next Pope. 

Contemporary prediction markets have their roots in modern economics, statistics, mechanism design, and computer science. 

Caltech’s Charles R. Plott and Shyam Sunder of Yale University introduced formal academic frameworks for them in the 1980s, and the first “modern” instance – the Iowa Electronic Markets – launched soon afterwards.

The mechanics are as follows. 

A prediction market for “Will Sam Darnold, the Seahawks’ quarterback, pass from inside the one-yard line?” would center around a contract that pays, say, $1 per unit if such a pass occurs. 

With people trading the asset back and forth, the market price can be interpreted as a probability: an estimate of traders’ aggregate belief about the outcome. 

A market price per unit of $0.50 implies a 50/50 probability.

You will buy if you think it’s more than 50% likely – say, 67% – that Darnold makes such a pass, and if he does, you will gross $0.67 for a price of $0.50. 

In the meantime, your purchase will push the market price and associated probability estimate upwards, reflecting the idea that someone thought the market was underestimating the likelihood. 

(And of course, the reverse is true if someone thinks the market is overestimating.)

When prediction markets work well, they can have significant benefits relative to other forecasting methods. 

Polls and surveys just give an opinion share. To convert that into a probability estimate, you must reason statistically about how the share you measured relates to the overall population. 

Polls also typically reflect just a snapshot in time, whereas prediction markets update as new participants and/or information arrive.

Crucially, prediction markets are incentivized. 

Buyers and sellers have “skin in the game.” 

They must think carefully about what information they have, risking their capital on the issues where they believe they are most informed. 

And the opportunity to leverage information and expertise in prediction markets can create incentives for people to learn more about an issue.

Finally, prediction markets have a big advantage in the breadth of coverage they offer. 

While someone with knowledge about events that may affect petroleum demand can go short or long on oil, there are plenty of outcomes we might want to predict that are not well supported by large-scale commodities or equities. 

For example, prediction markets have recently sprung up to try to aggregate estimates of when specific math problems will be solved – which is important for scientific progress (and a benchmark for AI).

Still, much is required for prediction markets to fulfill their promise. 

There will always be important market-infrastructure questions like how to validate and reach consensus on whether a given event has occurred, and how to ensure that the market’s operations are transparent and auditable.

There are also market-design challenges. 

For example, participants with relevant information must show up. 

If everyone is uninformed, the market’s price signal tells us little. 

Equally, people with all different types of relevant information must decide to participate, or else the prediction market’s estimate will be biased (as arguably happened before the United Kingdom’s Brexit referendum).

But if someone with perfect information shows up – for example, the offensive coordinator for the Seahawks will know if Darnold will throw from the one-yard line – that can also be a problem, especially if that person can affect what happens. 

If prospective participants believe that insiders will be trading in the market, they might rationally choose to stay away, causing the market to unravel.

Finally, there’s also a possibility that people might try to turn prediction markets from tools for aggregating beliefs into tools for manipulating them. 

If a candidate’s communications team wants the world to think they will win an election, they could use part of their war chest to try to sway prediction markets. 

That said, prediction markets are somewhat self-correcting in this regard, because people can always take the other side of a contract that pushes the probability estimate beyond belief.

Given the risks, prediction-market platforms must work to ensure greater transparency and clarity about how they manage participation, contract design, and operations. 

If they can successfully solve these puzzles, we can predict that they will continue to play a growing role in the future of forecasting.

Scott Duke Kominers, Professor of Business Administration in the Entrepreneurial Management Unit at Harvard Business School, is a faculty affiliate of the Harvard Department of Economics, Co-Principal Investigator of the Harvard Crypto, Fintech, and Web3 Lab, and a research partner at a16z crypto. He is the co-author (with Steve Kaczynski) of The Everything Token: How NFTs and Web3 Will Transform the Way We Buy, Sell, and Create (Portfolio, 2024).