More thoughts on predictive markets

From Patrick Boyle's recent video overview of prediction markets.

If you are a retail trader who’s tired of losing money to hedge fund algorithms, you might be tempted to look for an edge.

In traditional financial markets, acquiring non-public material information and trading on it will generally result in a conversation with the authorities and eventually a lengthy stay in a federal facility.

In prediction markets, however, insider trading is often described by proponents as a feature, not a bug.

That’s right—a feature.

If we look at recent events, this feature appears to be working remarkably well.

Last summer, a Polymarket user operating under the pseudonym Rico Suave 666 made a series of highly precise and highly lucrative bets regarding the exact timing of military strikes in the Middle East.

It later turned out that Rico Suave 666 was not just a very astute reader of geopolitical tea leaves. The Israeli government arrested two men, including an army reservist, for allegedly placing bets using classified military intelligence.

So to be clear, a soldier with advance knowledge of when bombs were going to be dropped used that information to win money on what is essentially a crypto gambling website, which is not really what people have in mind when they talk about the wisdom of crowds.

 One of the most annoying but absolutely on-brand comments to come out of the prediction market industry is that they are creating "truth machines." That combination of the pretentious and the portentous is exactly what we've come to expect from tech visionaries, along with the inevitable moderately dishonest framing.

Truth is a big word that brings with it volumes of connotations, used here in the narrowest possible sense to mean an estimate is more accurate (given that a large part of the blog's readership consists of statistics professors, I should probably acknowledge that even that grossly oversimplifies what we're talking about here).

"Truth" also puts a massive thumb on the rhetorical scales. Who could be against truth? Hell, it's part of Superman's credo, for crying out loud.

But the dishonesty really comes to a head when the people behind these markets argue in defense of insider trading. The whole point of the magic of the market is that it brings together a large number of guesses that have been weighted according to strength of belief. (There's also a good chance that some of the people reading this are economics professors, so, once again, I apologize for this sub-layman's summary.)

If one of the investors/bettors knows the outcome in advance, it's no longer magic; it's the cheap trickery of a fake psychic checking the mark's wallet while he's in the bathroom.

Insider trading is not itself a social good for at least a few reasons. For starters, the very concept of being an insider is heavily correlated with being someone with power, connections, and money. Allowing people to profit off of that status is, with respect to investors, effectively a tax collected from the have-nots and paid to the haves.

But it gets worse. Having mechanisms in place where inside information is valuable creates an incentive to create more inside information, or, put another way, to keep information on the inside that would normally get out. Let's take the example of weather forecasting, one of those cases where there is a clear social good associated with knowing the answer in advance. If a researcher with NOAA can get rich by plugging non-publicly available data into their models, they have every reason to keep as much data as possible out of the public domain. The net effect of this would be less, not more, "truth." Making data publicly available allows for more calculations, more research, more models. Our ability to make predictions based on government data which, in theory, belongs to all citizens is diminished, not increased.

This gets at the absurdity of the pro-insider argument: it doesn't even make sense on its own terms. Compared with other ways of getting information out there, insider trading is perhaps the least effective way of making prediction markets more accurate. In all but the most thinly traded of markets, insider information will only slightly move the needle, while by comparison taking that same information and releasing it publicly through the press or through the government would have a far greater impact on the value of those trades. In other words, insider trading only improves predictive power if the information doesn't come out through other means. Otherwise, it actually reduces the level of "truth."

Add to that the potential for gaming the system, either trying to influence events to make a bet pay off or trying to influence the prediction markets to game events.

The people at Kalshi and Polymarket argue for insider trading because they have to, because they have no real way of stopping it, and if people have to acknowledge how terrible the practice is, the only logical next step is strict regulation.

Finally, and perhaps most importantly, we have to ask ourselves what the social good of accurate predictions is in specific contexts. There are cases where knowing the future is a wonderful thing. We previously mentioned weather. This is an area where an accurate forecast is beneficial in any number of ways, from safety to resource allocation to infrastructure. Here in the West, knowing when to top off the reservoirs and when to plan for wildfire conditions would be of tremendous value. On the East and Gulf Coasts, having advanced warnings of hurricanes might be even more important.

But there are any number of events where "truth," in the sense of knowing the outcome in advance, is of no social good whatsoever. Sporting events obviously fall into this category. I would argue that general elections do as well, though I might get some pushback on this one. Just to be clear, this objection includes both polls and prediction markets, though the social costs associated with the latter are considerably greater. While there is a social good associated with polling in a primary—it can help party members converge on a candidate who is both acceptable and electable—all that prediction models do in general elections is tell the public who the public is about to choose for an office. (We'll get back to that one.)

Prediction markets, humanoids, LLMs.  It's the same scam. The boosters pitch their shiny new innovation has a magical machine that can do anything or tell us anything and we're all supposed to be so entranced that we don't ask inconvenient questions about how good a job it does or how it works or what it costs.