Why Prediction Markets are Cool

#FIVE HUNDRED YEARS IN ONE PARAGRAPH

In 1503 someone in Rome bet money on who the next pope would be, and a historian of papal elections notes the practice was already old by then. In 1591 Gregory XIV banned it on pain of excommunication. In 1907 Francis Galton went to a country fair to prove crowds were idiots and accidentally proved the opposite. From 1868 to 1940 the United States ran enormous public betting markets on presidential elections, quoted on the curb outside the New York Stock Exchange and printed in the Times as news, and then killed them. In 1988 some economists in an Iowa City bar rebuilt one as a toy. In 2003 DARPA built one and Congress killed it in twenty-four hours. In 2024 a federal judge finally said yes, and $3.3 billion moved through a single market on a single election.

In five centuries of people trying to shut these things down, not once was the objection that the price was wrong.

#THE PEOPLE WHO NEVER PAY

Someone will go on television tonight and say the Fed cuts in September.

It costs them nothing to be wrong. Nobody clips the tape, nobody keeps score, and if you hedge with enough conditionals you were right either way. Next quarter they book you again.

We've built a whole profession on this. Analysts, strategists, columnists, consultants, thousands of people paid well to say what happens next, and almost none of them ever settle up. They're paid to be interesting. Being right is optional. Being wrong is free.

He is not the problem. He is the incentive working exactly as designed. Nobody is paying for the pick, they're paying for the segment, and the segment is just as good when the pick is wrong.

We don't have a forecasting shortage. We're drowning in forecasts. What we don't have is a bill.

#A MARKET DOES THREE THINGS

Put a price on the future and let anyone trade it. Three things happen, and each one fixes something nothing else fixes.

You put a number on it. You back the number with money. It settles against reality, and reality doesn't take your call. No conditional to hide behind, no hedge, no quiet rewrite afterward. That one constraint is the whole difference between a forecast and a confession.

This is Hayek's argument from 1945 and it's the deep one. The knowledge a society runs on is never sitting anywhere in one piece. It's scattered across thousands of people in fragments, most of it tacit, a lot of it stuff the person holding it couldn't explain to you if you asked them directly. No planner can assemble that. No survey gets at it. A price does, because a price is the one signal that squashes everything everyone knows into a single number.

Galton found the mechanism by accident. He was eighty-five, at a livestock fair in Plymouth, and he collected 787 guesses of an ox's weight fully expecting to document mass stupidity. The median guess came in at 1,207 pounds. The ox weighed 1,198.

Nine pounds off. Not one person in that crowd got it right, and together they were within three quarters of one percent.

That's aggregation with no incentive attached at all. Now pay people to be right, let them trade as news breaks, and let anyone who spots an error take the other side and get paid for it. At that point you've stopped running a poll and started building an instrument.

Not once an election cycle. Not once an earnings call. Not whenever a committee finishes deliberating and a press office picks a release date. The second somebody who knows something acts on it, the number moves, and the number moving is the news.

#THIS ISN'T A STORY ABOUT ELECTIONS

Elections are just where the cameras are. Hewlett-Packard let its own employees trade on the company's printer sales, and in about 75% of the cases where there was an official HP forecast to compare against, the market got closer to the truth than the corporate process did. Sales staff and supply chain managers knew things the forecast couldn't reach. Google built the biggest corporate market anyone knows about, and Ford, Siemens, Microsoft and Best Buy all ran their own versions. Then researchers pointed the same tool at science. When psychologists and economists traded contracts on which published studies would survive replication, the markets called it right for 73% of 104 studies, and 86% of the ones pulled from Nature and Science. A market made of academics knew which of its own field's famous results were fake, and said so out loud before anybody ran the experiment. Kalshi's contracts on CPI, payrolls and Fed decisions now match or beat the Bloomberg economist surveys they compete against. Printers, papers, payrolls. The mechanism doesn't care what the question is about.

#THE SCOREBOARD

The Iowa Electronic Markets launched in 1988 with a five hundred dollar position cap, which regulators tolerated strictly as an academic curiosity. A toy, run out of a business school, with less money riding on it than a used car.

Across five presidential cycles and 964 head-to-head comparisons against public polls, the toy landed closer to the real result 74% of the time. It won by the widest margin more than a hundred days out, which is exactly when polls are measuring name recognition and a market is already doing arithmetic.

Wall Street's version got eleven of twelve presidential elections right between 1884 and 1940. No pollsters, no voter files, no models, nobody weighting a sample by education. Just prices, set by people who had to pay for being wrong.

But the one that should end the argument came out in February 2026, when the Federal Reserve published a staff paper on Kalshi. Not a think piece. A benchmarking exercise, run by an economist at the Fed Board with co-authors at Northwestern and Johns Hopkins, holding a prediction market up against the forecasts the Fed actually uses.

On headline CPI, Kalshi's median forecast came in with a mean absolute error of 0.063 against the Bloomberg consensus at 0.081, and the gap is statistically significant. On the fed funds decision itself, measured on the day of the FOMC, Kalshi's median and mode had a mean absolute error of exactly zero, against fed funds futures at 0.010. They point at September 2024, when the market was split between a 25 and a 50 basis point cut, and Kalshi put the heavier weight on the 50 that actually landed.

The sentence in that paper worth sitting with is the flattest one in it. "In no case is Kalshi significantly worse than the Bloomberg consensus."

That's the Fed's own research staff, comparing the professional forecasting apparatus against a website where anyone with an opinion can take the other side, and reporting that the website is at worst a tie.

#NOBODY CARES WHO YOU ARE

The part I like most about a market is that it's completely indifferent to credentials.

A trader with no followers, no byline and no green room access can move a price by being right more often than everyone else. The track record is public, permanent, and denominated in dollars instead of reach. Compare that to how we currently pick who to believe about the future, which is mostly pedigree, charisma, and how confidently somebody can deliver a sentence.

• ▸Being wrong costs you immediately and measurably, so bad forecasters get pruned out on their own and nobody has to run a performance review
• ▸Anyone holding real information, an industry insider, a local reporter, someone who just read the primary source, gets paid for it without building an audience first
• ▸You get a probability instead of a narrative. Not "experts are concerned about a rate cut" but 34 cents, which is a thing you can plan against
• ▸And it prices the questions nobody else will touch, because there's no institution whose job it is to forecast most of what actually matters

#IT ISN'T MAGIC

A thin market is a rumor with a decimal point. If a few thousand dollars moves the line, the price is telling you about one trader and not about the world, and thin is exactly what these markets are on a lot of the questions you'd most want answered. US regulation is still genuinely unsettled. What a federally designated exchange can list and what a state gaming commission thinks is an illegal bet are two live arguments running at the same time. And a market is only ever as good as the question it settles on, because a sloppily drafted contract will hand you a confident, precise, worthless number.

None of that is a reason to look away. It's a reason to read the question as carefully as you read the price.

#ONE MORE THING

A market gives you a number. It doesn't tell you why the number moved, which of the four thousand live markets is mispriced right now, or what you should do about it. It won't read the filing, notice that two related contracts contradict each other, or sit up at 3am watching a book for the moment the price breaks away from the news.

The instrument exists now. The layer that makes it usable doesn't.

That's Kosmos, the intelligence and agentic layer for prediction markets. Intelligence: the research, the signals, the reason a price moved, sitting right next to the price. Agentic: software that watches every book continuously, finds the mispricing and acts on it, at a speed and scale nobody sitting at a terminal can match.

And this is the only market I know of where that actually hands the advantage to the individual. You can't out-spend anyone into knowing whether a manager benches his starter in a dead rubber. Susquehanna figured this out early. They stood up the first dedicated prediction markets desk in 2023, they're the flagship market maker on Kalshi, and their answer to the opportunity was to go hire a sports trader. DRW and Jump are doing the same. The sharpest quant shops on earth looked at this and concluded the scarce input isn't capital or colocation. It's somebody who actually knows the thing.

Which is the whole opening. The obsessive who has read every filing in one boring sector, or watched one league for fifteen years, is holding the exact input a billion dollar desk has to post a job listing to get. What they never had was leverage: no team, no infrastructure, no way to watch four thousand markets at 3am. Agents are that leverage. They run the workflow, hold the whole book in view, and never get tired.

For five hundred years the edge belonged to whoever knew the most. For the last forty it belonged to whoever could afford the most. Prediction markets hand it back, and agents are how one person picks it up.

Five hundred years to build a working instrument for the truth. It's finally on, it's finally legal, and it's still sitting there mostly unread, because nobody built the thing that points it at something.

It'd be a damn shame and a waste if we don't capitalize.