Prediction Markets Are Not Good Markets (Yet)

Are prediction markets good? Rationalists and economists like Robin Hanson have been extolling their virtues for decades, and their promise appears closer than ever to being realized.

It’s certainly been interesting to read about the theoretical virtues of prediction markets from 2015-2020 and then witness them come into being and become a massive cultural phenomenon in 2025/26.

(To quickly establish my bona fides on this, I was waiting for Augur to launch in 2017 and excited when it did go live in 2018. In 2023 I started paying very close attention to Polymarket. I was a big advocate of Polymarket’s triumph over MSM in the 2024 election. I’ve been hopeful about prediction markets for a long time.)

They are certainly useful to me. Prediction markets aren’t necessarily more accurate than other markets, but they can do certain things very efficiently – compress vast arrays of noisy data into a single probability, and provide a real-time informational flow that is otherwise gatekept by governments, pollsters or news organizations. For people following elections or geopolitical events in real time, they are far better than the alternatives. Watching the 2024 election on the real time state-by-state view on Polymarket, it was obvious to me that Trump was going to win hours before any major news organization made a call. It’s just a superior informational experience.

However, the full promise of prediction markets has not been realized yet. The early believers envisioned a world in which:

• Corporate entities could financially hedge outcomes that mattered to them via prediction markets

• Markets could actually guide policy, through mechanisms like Futarchy

• Private information could be surfaced to the general public through the market mechanism, creating a kind of informational surplus

Neither the first nor the second thing are meaningfully real yet, and the third thing – arguably the greatest promise of prediction markets – is inherently paradoxical and coexists uneasily with our established legal framework and prudent market design. Mostly, prediction markets are used today as a regulatory arbitrage to get around restrictions on sports betting, and a tool for retail to bet on pop culture and politics. (I am not the first to express these misgivings – see Vitalik and Agustin Lebron.)¹

Don’t get me wrong. I am fascinated by prediction markets. I have been waiting for them to exist ever since I read about Truthcoin and Augur a decade ago. I now consume Polymarket data every day and consider it an indispensable part of my media diet. I’m just not convinced they are actually going to work in the way early advocates intended.

There are two main structural problems that I see with prediction markets:

1. They aren’t useful for hedging, and may never be, due to market fragmentation and a lack of natural buyers on both sides of the trade

2. Part of the value of prediction markets lies in the surfacing of insider information, which is and will likely remain illegal, and will cause trader disillusionment with time

Prediction markets have other contingent or non-structural problems, which can indeed be solved. For instance, Polymarket relies on easily-manipulable resolution oracles provided by UMA. They need to get rid of this. Both Polymarket and Kalshi are exposed to legal and political risk in the US. But I’m more interested in the structural factors.

Prediction markets aren’t good for hedging

I’m hardly the first person to point this out, but it bears repeating.

Crypto people sometimes just assume that if you put the market structure in place, a liquid two-sided market will magically appear. This is a reasonable delusion to hold, because crypto markets did appear out of thin air, and crypto involves the trading of some of the most worthless tokens ever devised. But real, sustained markets need a fundamental reason to exist.

Take derivatives. Derivatives don’t exist “because people want to bet on the price of some underlying commodity”. They exist because there is a massive underlying market and someone has inventory risk in that market, and someone else is willing to absorb it for a price.

More concretely, take wheat derivatives, probably the oldest market in the world (dating back to Mesopotamia in 2000 BC). These are good markets because some entities are naturally long and some entities are naturally short. Producers like farmers are naturally long, and want to hedge. They know they will produce a certain amount of wheat by a certain date, and they might want to hedge to lock in a guaranteed margin. On the other side, there are entities like cereal manufacturers and livestock producers that are naturally short – they know what they have to buy for the next few quarters, and might want to lock in a fixed price. You can find natural hedgers on both sides for a variety of derivative markets:

• Buyers of oil shorts: producers, shale operators

• Buyers of oil longs: airlines, shipping companies, refineries, chemical manufacturers

• Buyers of FX shorts: importers, corporations with offshore liabilities, governments with foreign FX debt

• Buyers of FX longs: exporters, multinationals with foreign revenue, foreign investors in local assets

• Buyers of interest rate shorts: corporations with floating rate debt, mortgage lenders, leveraged borrowers (they lose if rates rise)

• Buyers of interest rate longs: pension funds, insurance companies (they lose if rates fall)

These derivative markets exist because these economic actors have real, offsetting exposures. Producers have inventory risk. Predictable consumers have spot risk.

Now let’s consider prediction markets.

You can see in theory why they’d be useful for hedging. Many economic actors have exposures to elections. The long side of the Trump trade was something like energy companies, defense, crypto, AI, rare earths, private prisons, and so on. Prediction markets can be as specific as you like. A biotech company might want to hedge out some of their exposure to a drug going through FDA trials by shorting it on a prediction market. Multinationals might want to bet on granular markets covering tariffs or war or other political shocks. Apple or TSMC might want to bet on an invasion of Taiwan.

But in practice, prediction markets aren’t good ways to hedge.

The first reason why is that the more useful the market is to the hedger, the less liquid it is likely to be. The big obvious problem with PMs is fragmentation. Markets are often fragmented across a range of maturities (USA to invade Iran by Date X, Date Y, etc etc) or formulations “USA to Strike Iran…” “USA to invade Iran…” “Regime change in Iran”.

A corporate hedger probably doesn’t care all that much about “Trump wins” or “Pro-business environment in 2026” in the general sense. They care about specific things:

• Tariff policy concerning their industry

• FDA approvals for their specific drugs

• The FTC approving a specific merger

• Corporate tax policy in a specific state

And so on. But the more specialized and bespoke the market, the less likely there is to be a general interest in betting on it, among the speculator population. A lot of people wanted to place bets on the 2024 election. But not a lot of people want to take the other side of Trump wins -> FDA adopts a more permissive stance -> drug xyz gets approved or any other such unique permutation. So almost by definition, the markets that hedgers care most about are bound to be more bespoke and hence less interesting to traders in general – and thus less liquid.

Commodity derivatives don’t have this problem. If you are a farmer and anxious about locking in your returns for the season, you can sell contracts corresponding to the exact number of units that you expect to produce. But a corporate hedger trying to manage their exposure to some political development needs to pinpoint it exactly, rather than taking a vague bet on a broader but more liquid market.

In technical terms, this is basis risk – the idea that your hedge may not correlate perfectly with your exposure. A wheat derivative has minuscule real basis risk – if you expect to harvest 10,000 bushels and you sell 10,000 futures, you are tightly hedged to the underlying. If you are a vaccine manufacturer, and you want to hedge the risk of the FDA shutting off vaccine programs, you might want to bet on that specific market, but instead be forced to buy shares of TrumpWin – something that imperfectly correlates to your exposure. So there seems to be an inverse relationship between the liquidity and generality of a market, and its specific utility in hedging an event for the seller.

In addition to this problem, for the hedge to work at all, you need someone who is willing to long the market that the corporate wants to sell in size. Who are the traders wanting to long “merger between company A and B is consummated” or “tariffs on Korean cosmetics lifted by July 2027”?

This brings us to our second problem. For the “corporate hedger” concept to work well, you need an army of speculative noise traders to effectively “subsidize” these shorts by taking the other side of the trade. But these hedge markets are unbalanced by their very nature.

Imagine the case of the biotech firm wanting to hedge their exposure to a drug approval. Consider the status quo. The company is naturally long approval. The employees, management, and shareholders may want to reduce their risk by buying “No” shares on a prediction market. So there’s strong demand on the short side. Now where does the offsetting long demand come from? Everyone who cares about the drug is presumably long via the stock. So you are left with unaffiliated traders who want to bet on approval but for some reason don’t want to hold the stock. This is an unbalanced market.

Now consider a solar manufacturer which may want to hedge an adverse Supreme Court decision striking down solar subsidies. They might go long YES on that adverse judgment. But who is on the other side of that? Someone that is naturally long non-renewables presumably, but do they really care about the Supreme Court decision? This is a case where a catalyst affects one side of the market a lot more than the other. Asymmetries like that don’t lead to good markets.

Now a smart reader might think to themselves, “well, insurance is asymmetric. No one wants to be naturally long “good weather in Florida”, and yet there are companies that will sell you hurricane insurance. So clearly, asymmetry alone isn’t enough to eliminate the usefulness of a market.” And indeed, this is true. There are all kinds of hedges available where the seller simply wants to collect the premium.

The difference is that insurance markets are written on losses that are objectively measurable and tightly specified. A homeowner insures a defined asset for a defined value against a defined risk. A CDS contract references a specific bond with a legally defined credit event. The payout maps directly to the economic loss. The ones writing the options or selling the insurance have large balance sheets and diversify the risk across many independent exposures. By contrast, a prediction market on a Supreme Court decision or regulatory change is a coarse, binary contract written on a political event whose economic consequences are heterogeneous and path dependent. The solar manufacturer does not suffer a $100 million loss simply because the court rules one way; the impact depends on implementation, legislative response, further rulings, and so on. The problem is not asymmetry per se, but the basis risk (the drift in relationship between the exposure and the hedge value).

One obvious conclusion here is that useful corporate hedges can be viable and already do exist – but these are already wrapped inside the form factor of equity options, commodity derivatives, insurance or CDS. The promise of prediction markets is that organically, markets of sufficient liquidity and breadth might develop naturally such that corporates could come to rely on them for otherwise bespoke hedges. But this seems unrealistic. The markets that seem to appeal to speculators have virtually nothing to do with the markets that hedgers want to buy. And these problems seem structural, not contingent.

Another way of putting this is: why would you expect a retail trader to be better at or more willing to write an insurance policy than a professional actuary with a large balance sheet behind him? And if your answer is “that retail trader might write a cheaper insurance policy,” then the market cannot last – as retail bankrolls will go to 0 in time.

Insider trading is both the point and the death of prediction markets

There are three types of entities active in prediction markets. As above, there are some corporate hedgers. These are the best entities to have because they are “natural buyers” of the product. Such is their motivation to trade, they might be willing to cross the book and buy regardless of market conditions. For PMs to work in the long term, you need to attract some of these, otherwise the markets will die out.

Then you have noise traders. These are the vast majority of users on prediction markets. They are trading for entertainment value. If they weren’t trading on PMs they would be trading 0DTE options, Fanduel/Draftkings, or trading memecoins. These are the people Kalshi and Polymarket try to target with their ad campaigns. Noise traders generally lose their bankroll over time (both through fees and because they are trading against informed flow).

Lastly, you have informed flow. Entities that are professional traders and have specific knowledge of a certain niche, or have literal inside information within their organization that they are monetizing by trading against noise traders. (I’ll ignore arbitrageurs and market markets for now).

The critical problem for prediction markets is this:

The social value of prediction markets derives from financially incentivizing insiders to divulge confidential information, but this collapses noise trader confidence in the market over time.

This isn’t some big secret. Prediction market theorists like Robin Hanson² have long explained that there’s a tradeoff between PM utility and insider trading prohibitions. In 2024, he told Decrypt:

If the point of [prediction] markets is to get accurate information on the prices, then you definitely want to allow insiders to trade, even if that discourages other people from betting because that makes the prices more accurate. And that’s the priority.

This is actually a perfectly coherent viewpoint, but the major prediction markets don’t acknowledge it (for obvious reasons). Kalshi bans trading on insider knowledge in their ToS. Polymarket doesn’t explicitly ban insider trading, but has a catch-all prohibition against violating laws generally.

(A quick note here: I’m using insider trading to mean “misappropriating organizational information that is not owned by you to obtain a financial return”, rather than the more specific, securities-focused definition. In practice, if you work in government and make money by selling military secrets (to a prediction market), you have broken the law, but you haven’t engaged in “insider trading” in the sense of the SEC and public securities markets.)

People won’t trade in markets they perceive as unfair

The core tension of these markets is that they are informationally useful to the general public if they can actually persuade insiders to reveal corporate or governmental secrets for compensation. That way, the public gets knowledge that may otherwise never have surfaced. If prediction markets can consistently surface information that is legally prohibited in comparable securities markets, they might come to be seen as more informationally reliable.

The problem is that this quality trades off against market integrity. Securities regulators go to great lengths to ban trading on insider information in public markets, because non-insiders wouldn’t risk trading in those markets if they thought they were up against insiders all the time. The social utility of the securities markets would collapse: they would become less liquid and the cost of capital for public firms would rise. Similarly, sports leagues try and uphold the integrity of the game by banning athletes from betting on their own games (if they didn’t, fans would think it was rigged, and ticket sales would drop).

If it becomes known that prediction markets are rife with insiders monetizing proprietary information, the general public will become disillusioned with them and stop trading on them (or limit their trading only to markets that are largely immune to insiders, like sports, or the weather).

Already, prediction markets are developing a reputation for being a dumping ground for insider info. Many people don’t realize that it’s generally illegal to misappropriate organizational knowledge for personal gain even outside of a securities context, so there are many examples of this, and certainly many more to come this year. Unusual activity suggestive of insider trading has been spotted in markets like:

• The 2025 Nobel Peace Prize

• Google’s 2025 Year in Search Rankings

• Maduro’s capture in early 2026

• Israel’s 2024 strikes on Iran

Of course, it’s possible that these bets were made by smart outsiders rather than insiders. There’s nothing wrong with looking at OSINT flight data and correctly predicting a US war with Iran. There is something morally and legally wrong with a military staffer receiving a briefing about an imminent attack and trading on it. No one knows for sure, but the fact remains: the incentive is there, and because Polymarket in particular is on chain and pseudonymous, it’s possible to (sort of) cover your tracks.

If prediction markets develop a reputation as an “anything goes” zone to monetize insider information – and they seem to be going that direction – smart non-insider traders will become disillusioned and pare back their trading activity. It’s just not worth trading if the guy on the other side of your trade knows something you don’t. If you can’t spot the sucker at the table…

Informed flow will harvest noise traders until they go bankrupt

Markets have a defined ecology to them.³ They are composed of different types of entities. Let’s return to our taxonomy. We have the hedgers, the noise traders, and the informed flow.

The unspoken feedback loop that prediction markets are relying on is something like the following:

If we can attract enough noise traders, informed flow will emerge to harvest them, creating markets of high informational quality, which will in turn further attract traders. Over time, the markets will become sufficiently liquid that natural buyers (corporate hedgers) will find them useful, creating a sustainable market.

This explains why Polymarket and Kalshi advertise to the general public so aggressively. They presumably realize that they need to get a critical mass of noise traders, who attract the informed traders, who attract the hedgers. Noise traders are the plankton of the prediction market system. The whole ecosystem relies on them. The risk is that they get farmed by the informed traders too quickly or lose faith in the fairness of the market. If they go, the whole ecosystem collapses.

Consider a couple different markets. Online poker in around 2005 was a magical place. I like many others now in crypto got my start there (and my first taste of an exchange insolvency in 2011). The sector was awash with unsophisticated recreational players and they were happily farmed by the pros. After the DoJ shut down easy access to the main online poker sites on Black Friday in 2011, it became more challenging for US users to play online.⁴ The big influx of cash from US retail stopped, and the games turned into grinders playing against other grinders. Sophisticated bots emerged. Any remaining noise traders were relentlessly farmed by the bots and the pros, and their bankrolls went to 0. The space basically disappeared. The reason was that there was no “natural buyer” once the easy money retail guys went BK or were shut out by regulators.

Now think about the memecoin market.

The second big memecoin era was in 2023-2024, defined by retail participation. Fortunes were made and pump.fun made people feel like they were playing a relatively fair game. Pump standardized the launch process. It became trivial for anyone to launch a coin. No presale, private team allocation, no whitelist, etc. Price set by a transparent bonding curve. The game was simply predicting what memes would hit and which wouldn’t.

However, all was not well under the hood. Pump fun memecoins only seemed fair. They weren’t. Over time, sniper bots and MEV emerged. More sophisticated users were able to farm the noise traders who were under the impression that they were playing a fair game. Swiftly rugged influencer coins in 2023 and 2024 created a state of exhaustion. Devs began sniping their own launches. They created the illusion of decentralized ownership by splitting their accounts into multiple wallets. Telegram and Discord groups created layers of coordinated pumps designed to ensnare onlookers.

Memecoins died because noise traders got farmed by the pros, and they lost confidence in the integrity of the market. Major debacles like the Milei/Libra and Melania coin launches revealed to the world that the memecoin market had been warped from something light-hearted and playful to an industrialized extraction scheme run by grifters like Hayden Davis. In addition to being out of cash, retail traders suffered a total confidence collapse. There of course are no natural buyers of memecoins, so once noise traders leave, the game is over.

If this guy was my apex predator, I’d probably quit too

Prediction markets have more social value than memecoins or online poker. That much is indisputable. But they face the same risks. Absent any permanent natural buyers – and I’m just not convinced the hedgers will ever find a use for these markets – it’s going to be a matter of informed flow farming noise traders. Kalshi and Polymarket need to grow their userbase faster than the noise traders lose money to the smart money.

For now the strategy is working, but that’s because the prediction markets have found a regulatory loophole: they can offer sports gambling (for the moment) without facing state-by-state restrictions that ordinary bookies have to deal with. (There’s also a more complex tax arbitrage reason.) From the dashboards I could find, it looks like sports composes 90% of Kalshi volume, and around 40% of Polymarket volume (followed by crypto and politics).

Source: DataDashboards on Dune

Source: The Block

Sports betting is keeping prediction markets alive. They could build sustainable businesses simply by cannibalizing existing bookmakers under a new more permissive regulatory model. Sports is less exposed to trader disillusionment from insider trading, since sports outcomes are more stochastic and less manipulable.

Outside of sports though, serious risk remains that insider trading scandals (and convictions, like that of the Israeli reservist) will give retail traders the impression that markets are rigged, causing them to desert the platforms. I predict that this year, a flurry of insider trading episodes will convince the platforms to dramatically beef up market surveillance, and cause Polymarket in particular to move away from the pseudonymous model. Though I don’t think prediction markets will ever realize the dreams of Robin Hanson and the rationalists, they might still earn a tidy living with sports and cultural markets. I just don’t think they will ever be much more than that.

1. A lot of my thinking on this was influenced by Agustin. Please read his piece as well!

2. If this article seems like a long criticism of Robin Hanson, it’s not. He’s expressed his own misgivings about the public state of prediction markets. I think he’s a genius, in fact.

3. I recommend reading Andrew Lo’s adaptive markets if you want to learn more about this

4. In fact, my first taste of Bitcoin came from trying to access one of these offshore casinos