By Alan Wu • Published March 24, 2026
Originally posted on X.
The next phase of prediction market growth will not come from more degen products. It will come from markets that are useful to people who never trade them: for better public signals, hedging where traditional systems fail, and accountability.
A publicly observable prediction market price is a public good. Once a market prices a recession by Q3 at 70 cents, anyone can act on that signal. The signal is non-rivalrous and non-excludable. In theory, providing a public good efficiently requires accounting for every beneficiary's willingness to pay. That never happens. The people who benefit most from accurate estimates are often not the ones trading.
This matters because not all prediction markets can pay for themselves. Some markets can support good liquidity on their own. Others are harder. When adverse selection risk gets too high, keeping spreads tight enough for the market to be meaningful becomes too costly. The market either gets wider and less useful, or it needs explicit incentives to keep liquidity there. Some markets earn enough in platform fees to cover those incentives. Others do not.
That is why many socially useful markets never get meaningful liquidity. Think policy outcomes, scientific forecasts, and long-horizon geopolitical risk. These markets can affect millions of people, but the people who benefit from the signal are much broader than the trader base. The payoff is delayed, diffuse, and hard to trace back. A prediction market might help society make a better decision years in advance, but the downstream gains rarely flow back to the market that produced the signal. So the market never takes off, the signal never exists, and the benefit is never realized.
Some markets generate more in fees than they require in liquidity support. That excess can subsidize markets that are socially valuable but not commercially self-sustaining.
But cross-subsidization should be deliberate. That means being intentional about which markets get created and which ones get incentives. Novelty can buy short-term attention. The bigger prize is creating markets whose signals spread because they are broadly relevant, useful, and worth citing. That is how the category expands beyond the people who already know it exists.
This is the same logic newspapers use to fund investigative journalism with ads. The profitable side sustains the valuable side. Optimal pricing can involve below-cost service on one side, funded by surplus from the other.
Cross-subsidization is a platform-level strategy. But outsiders may also be willing to subsidize a market if they value the resulting public signal more than the cost.
Consider a politician who promises a major project will be delivered on time and within budget. An opposing politician or advocacy group could subsidize a prediction market on that claim. If the market sits at 6%, that number does more persuasive work than any ad. It is not rhetoric. It is information backed by capital.
More broadly, subsidizing a market is another way of paying for information production. If the resulting signal improves real decisions, the cost of supporting the market may be small relative to the decisions it helps shape.
The point is that someone can value the existence of a credible price more than the cost of creating the market. The subsidy is not always reliant on altruism.
For this ecosystem of motivated funders to actually emerge, the infra has to support it. First, platforms can cross-subsidize internally, directing fees from stronger markets toward useful ones that would not survive on their own.
Second, let anyone subsidize a market. Polymarket has already moved in this direction with their Sponsor Market Rewards feature, which lets any user deposit funds into a smart contract that automatically distributes rewards to liquidity providers on a specific market.
Another form of subsidy is just showing up with liquidity. Prediction market order books often get wide and gappy, which makes both trading and the signal worse. A platform can counter that by supplying passive, AMM-like liquidity. That is effectively a subsidy: the platform is paying to keep the market continuously tradeable and the price worth looking at.
Another direction is giving platform users a way to participate in market creation. This does not require a giant leap toward full permissionlessness, but it can meaningfully expand who gets to decide which markets deserve to exist. Hyperliquid's HIP-4 points toward democratized prediction market creation on shared infra rather than through a centralized gatekeeper. It also pairs that with a call-auction mechanism to help new markets bootstrap a better opening price.
There are still things to think through.
The default assumption is that a prediction market's value equals the accuracy of its price as a probability estimate. But market prices are not perfect probabilities. Structural frictions can push price away from average belief.
Some markets can be useful even without perfect calibration.
Consider someone buying YES on a hurricane hitting their region not because they have better info, but because they want to hedge. They are optimizing for expected utility, not expected value. Meanwhile, a counterparty with no exposure to the hurricane operates on pure expected value terms. Both sides are behaving optimally.
The price drifting away from probability could make the market look "wrong," but it does not make it useless. Either it gets arbed back, or the market settles into a state that reflects something beyond pure probability. In this case, it is supporting people whose insurance companies have pulled out of their area. The market's function has shifted from pure information provision to risk transfer, and it may be even more useful to the world in that mode.
There are also very few clean ways to hedge election outcomes, policy changes, or congressional actions anywhere else. If people and institutions with real exposure use these markets to hedge, it can attract larger market makers and sharper traders over time because real flow is worth competing for.
Even for highly improbable events, where direct hedging can look capital-inefficient, there are plausible designs that could make this kind of risk transfer more practical.
Most information sources we encounter come with some bias. People know this and discount information based on who is saying it. Useful markets can improve discourse quality by forcing specificity. Public claims usually arrive as floating facts, detached from the choices that produced them. Prediction markets surface those choices: the question has to be stated clearly, the resolution source has to be defined, and disagreement can be expressed in price rather than rhetoric.
In a world where AI makes words cheaper, capital-backed specificity becomes more valuable.
The point of subsidizing useful markets is not to fund them forever. It is to prove that they are worth having.
A policy market, a geopolitical market, a public health market: maybe it needs subsidy at first. But if the signal turns out to be useful, people notice. A journalist cites it. A think tank watches it. An agency realizes too late that the market saw something they missed internally. Once that happens, demand starts to form around the category.
This loop is real but leaky. People can benefit from a signal without ever tracing that value back to the market that produced it. Useful forecasts are basically something people only learn to value after they have seen one work in their own domain.
I have a simple dream here: when I am riding the subway, I would rather see probability estimates for real-world events than ketamine and Ozempic ads.
That is how the category gets bigger. Not because more people want to gamble, but because more people realize these markets are actually useful.
Grow the pie, and make its growth something the world actually benefits from. Prediction markets do not have to choose between growth and impact. The best version of the category makes them the same thing.
Thank you to @ryanchern, @distbit0, @The_Zhukeeper, and @tenad0me for thinking through this with me.