Prediction Markets and the Social Value of Information
There's appeal, but I don't think I like them very much
I would have liked prediction markets if I’d thought about them much before I saw them in the wild.
I’m a fan of markets. Why not learn what people really think by having them put money on the line? I do like forcing pundits to bet. We should tax bullshit. It has negative externalities.
But now, we have data, and, unfortunately, we have to use it.
Another argument I would have found convincing in the past: who cares if there’s insider trading? In fact, that will just make the markets better at predicting because all information, including that available to insiders, gets incorporated into the price.
The idea that information is valuable is central to economics and intuitive. The more information is reflected in prices, the better decisions we can make about capital allocation. The misallocation of capital is a major drag on productivity.
A few things were missing from my model of the world:
The value of information is heterogeneous.
The value of information does not correlate with market size.
Not all information is disperse and diverse, which is where markets shine.
There are, naturally, negative externalities to gambling. So, the market's social value does not start at 0.
For example, one of the larger prediction markets was a bet on who would win the New York City mayoral race. What is the value of this information before the election concludes? Perhaps it provides a means to hedge against a certain candidate’s effect on asset prices if he wins? Or, aggregate information from this market’s outcome can be used to trade in the asset market.
Markets do not create information. They reveal and aggregate the information people bring to them. So, what additional information is the market aggregating here? It can’t come from public polls, which are, well, already public. It does generate a numerical index of the “vibes,” which are public knowledge but difficult to quantify. It’s not clear how predictive the vibes are, though, and, in any case, they are already reflected directly in asset prices. From insiders, it incorporates private polling data, which is probably the most useful information generated by the market.
But couldn’t those insiders use that private polling data to directly trade on assets that will (allegedly) be affected by a certain candidate’s win? Then, that information would already be in asset prices.
No, is the answer, I think, but the reason for that is that political insiders are small in asset markets, so they can’t really move the price, but large in (some) prediction markets, given their current size.
[I have an armchair theory that media fascination with prediction markets is that polls report their results in terms of proportions, while prediction markets return results that give an implied probability of so-and-so winning. Of course, polls could also return probabilities of winning with some kind of Bayesian inference, but they don’t. An opportunity for a polling company.]
And therein lies the reality of the information value of prediction markets: it depends on people with meaningful information making up a sizeable enough percentage of the market that their private information gets reflected in prices. If the only people with meaningful information are insiders and insiders are small, then they earn profits by trading in the market because their bets don’t move the price, but we don’t learn any special information from the market that we don’t already know.
The prediction markets that would be most useful are not political horse races or sports gambling, but predictions about events where the information is difficult to quantify and aggregate into a belief in the abstract, and where information is diverse and widely distributed.
For example, the market for apples aggregates information on the value of apples. Everyone has their own apple preferences, so the information is very diverse. No one else knows your preferences, so the information is widely distributed. Markets aggregate all this diverse and dispersed information into a price for apples, which summarizes how apples trade off versus any number of other goods.
Markets like these are not popular topics on Kalshi. Look at the home page. I struggled to find topics where many individual people have meaningful, diverse information about them. I’ll discuss a few that were (arguably) close to that before jumping into what’s mostly there.
“Will Americans receive stimulus checks?” Of course: how would any non-insider have useful information to contribute to the market here? But, it does, I suppose, operate as a hedge of some kind in case the government has to borrow to write stimulus checks, and so it might offer some idea of how exposed people are to the demand for treasuries or something (which is diverse, dispersed information)… but there are already so many ways to do this that have a market size of more than $1.5M.
And, sure, there are some markets here that are more or less just typical financial instruments. There’s one directly betting on when Bitcoin will cross $100k again, which is just a typical financial market that aggregates individual demand information. But, again, this isn’t and wasn’t really the pitch for prediction markets. This is just a normal financial market.
“More tech layoffs in 2026 than 2025?” While there’s some alpha from being an insider, it’s limited by the fact that you only know your company, and the market can effectively aggregate all these signals about the tech industry more generally. Workers without direct knowledge of pending layoffs do have some information on them. You can usually tell when it’s coming, and such a market might be a hedge against the possibility you’re laid off. Others outside the tech companies might have some useful leading indicators worth aggregating as well, e.g., that it’s getting harder to close commercial real estate deals in the Bay Area.
The combined market size for these and a few other roughly information aggregative markets is dwarfed, of course, by the $293M market for “PGA Championship Winner,” which is what these sites are really about.
“For where your treasure is, there your heart will be also.”
If you write down a typical model for a prediction market, you’ll start saying things like “each individual gets a signal x…” but they don’t really. Not in most of these. As much as I’d like to believe I have some signal about what’s going to happen with geopolitics, I don’t. My beliefs on the likelihood of various outcomes of the war in Iran are just an index of public information. I have no new signal to offer the market.
So, then, what’s left? If we don’t really draw signals, then we’re left with non-insiders really just placing bets on the events because they are risk-loving, either because they are in a relatively flat part of the wealth utility space and need to get a great outcome to move up a level, or because they have “an innate preference for risk.”
Go to Kalshi and ask yourself: what price in each of these markets would cause me to enter? Most of these, within a very broad, reasonable range, I would have no reason to think are mispriced, even if the price were like +/- 20% from what it is currently. I’d just assume the market had more information than I did.
The same isn’t true in the stock market. For example, if the price of Apple were 20% lower than it actually is, I could look at their financials, public roadmap, and go, “Wait a second. This is wrong. I am going to make a boatload of money. Can I take out a second mortgage on the house?”
So, this is why I think prediction markets have failed in reaching their promise:
For most of the predictions we want the markets to make, relevant information is highly concentrated.
There are very good substitutes for the more meaningful markets that directly focus on more traditional financial transactions and don’t rely on people with “innate preferences for risk” to get liquidity.
I’m not sure that we should like a system that functions as a transfer payment to insiders from people who think the key to improving their lives is to put it all on black.
I’m not for banning prediction markets, of course. People should generally be free to do as they will. Many vices and antisocial behaviors are legal.
We still let people be Dodgers fans, among other things.
Thanks for reading!
Zach
Connect at: https://linkedin.com/in/zlflynn