Good job Kalshi!

The thing is, every two years, prediction markets are amazing. Every normal American wants to know one thing when the polls close on Election Day, which is “who won?” You can go on social media, and people who are abnormally addicted to politics will be posting stuff like “watch the results in XY-23, that was R +13.5 in ’24, if it’s within two points then the Rs are going to have a long night,” and you can be like “ah, yes, I will be sure to watch those results,” and then someone else will post “the latest early ballot drop in PQ-7 has way more shy NC WWCs than expected, it’s going to be a long night for the Ds,” and you’re like “ah I see,” and various news sites will be calling races with 0.3% of the votes counted, and there will be a vast mosaic of information that you can, with great effort and attention, piece together to get a picture of how the evening is going.

And if that’s fun for you, great. Clearly a lot of people consume this stuff as entertainment. But I personally am busy and nervous and like to go to bed early, and it would be nice if someone else would piece everything together, weight it appropriately by how much it matters, and generate one simple answer to the question “who won?”^[1] Obviously, 20 minutes after the polls close, that answer will not be definitive. But instead of me frantically looking at all the data points as they come in and updating my expectations about who won, someone else should do that and just tell me the most likely answer and the degree of confidence. Who? “Nate Silver” is a reasonable answer, but (1) it is hard to rely entirely on any one expert and (2) he has reasons of professional reputation not to overreact instantaneously to each new data point, whereas I want immediate gratification.

The best answer, these days, is “prediction markets,” Kalshi and Polymarket and so forth. There are a bunch of, I don’t know, proprietary trading firms and high-net-worth sharps and political analysts and degenerate gamblers trying to make money by betting on elections, and the evening of Election Day is their Super Bowl.^[2] If the results in XY-23 come in D +2.5, and your quantitative or gut-instinct model tells you that implies an 85% chance of the Democrats winning the Senate, and “Democrats control the Senate” is trading at 75%, then you go buy a ton of that contract until the price gets closer to its 85% fair value. You get money if you’re right, and I get something much more important, which is an early bedtime.

And so in fact I had Kalshi’s “Midterms Hub” up on my computer yesterday evening,^[3] and I looked at it periodically, and as the dribbles of information and misinformation and analysis and spin came in on X and Threads and Bluesky and television, I didn’t look at them. Kalshi traders did, and synthesized them, and by 8:30 they told me that there was a 98% chance that the Democrats would control the Senate, and I went to bed. I simply did not need to sit around gathering and interpreting the tea leaves from XY-23; I could outsource that to other people. They did great. Bloomberg’s Joe Weisenthal writes:

Unambiguously, there was no better way to understand the state of the race in real time than following the prediction markets. I had TV news on, switching between the networks, while also keeping track of the races on Kalshi and Polymarket, which forecast a blowout hours before the traditional media made that call. 

The blue tsunami thesis looked very strong by around 6:30 p.m., the moment we got the first batch of ballots from Indiana. The Democrat’s odds to win the night moved up everywhere very fast. Meanwhile, the TV anchors weren’t even close to making big calls.

There are good journalistic reasons not to overextrapolate from early data and to be cautious about making bold predictions. There are good gambling reasons to do the opposite. The gamblers were informative!

Here I am talking mostly about how good prediction markets were over a few hours yesterday, from say 5 p.m. to 9 p.m. New York time. They were also pretty good, you know, last month: Candidates that have been favored in prediction markets tended to win, Kalshi is generally well calibrated, etc. (Kalshi reports today that “the candidates traders were favoring won in 98% of [resolved] races,” measuring favorites as of 5 p.m. on Election Day; Kalshi tells me that the favorites as of last Tuesday won 96.9% of their called races.) Prediction markets give people financial incentives to examine polls, construct quantitative models, do door-knocking road trips, build AI image-processing tools to count candidate yard signs from satellite images, etc., to find an edge, and those incentives work.

But the trading dynamics on the evening of Election Day are a bit different, less “what is the probability of Candidate X winning” and more “has Candidate X already won but no one has noticed yet?” There is a premium on simply knowing what has already happened 10 minutes before everyone else does. In some abstract sense this race to know first is not that socially useful; no company is going to change its production decisions because it knows the election results 10 minutes earlier on a Tuesday night. But as a consumer product, man, it’s a good product. I really appreciated it.

Obviously, on the other 729 days per biennium, the prediction markets mostly offer a different product, sports gambling, which I, uh, discuss with occasional skepticism around here. But the election stuff is great! I love it! No complaints! I realize that people have complaints! People might insider trade on elections. People might manipulate elections (running fake candidates, sabotaging real candidates, doing fake polls, etc.) to make money in prediction markets. People might manipulate prediction markets (buying contracts uneconomically in large size) to influence elections: If you spend money to inflate a candidate’s market-implied probability of winning, maybe that will somehow (by driving donations, coverage, enthusiasm, etc.) actually make her win. There is something tawdry and cheapening about betting on elections. Sometimes the prediction markets are wrong. Etc.

These do not bug me that much. I get it! But it seems to me like our current equilibrium is that US midterm election prediction markets are big enough ($2.5 billion of volume on Kalshi) to provide real incentives for accurate predictions, but not big enough to, like, swamp the actual elections. US midterm elections have sufficient independent significance that no one’s really going to throw a Senate race to make a buck on Kalshi. Prediction markets trade predictions on all sorts of nonsense where the betting, in some sense, overshadows the event. Here the event is pretty important.

Everyone knows all this. Most prediction markets have their origin story in US elections. Every prediction-market entrepreneur wants to talk about (1) how good their markets are at calling elections and (2) all the fabulous use cases for hedging real business risk in election prediction markets. Prediction markets are “truth machines,” as the saying goes, and knowing the truth about elections seems pretty important and socially useful.

The big conceptual problem for prediction markets is that they create large, infrequent, positive externalities for the world, and have to figure out how to capture some of that value for themselves. Every two years, Kalshi tells you who will win the election with modest confidence a month out, and then with incredibly useful certainty like three hours before the election is called. The people providing that service — the people trading on Kalshi to make the probabilities right — make, in the aggregate, zero dollars, or a negative amount of money net of fees. Maybe that’s not quite right: Maybe there is a large group of negative-expectation Kalshi election traders who are just in it for fun, or who are hedging business risk from the election, or who are hedging the emotional risk of their favorite candidate losing, or what have you, and all of those traders are financially subsidizing the minority of sharp traders who are trying to make the probabilities right. The sharp traders are providing insurance and getting paid a fair rate for it. This is probably true in some sense, and crucially important to any financial market: You can’t really have a robust efficient financial market in which everyone is a sharp; the sharps have to be able to make money, in expectation, for the market to work.^[4]

Still, the point is that Kalshi election traders as a whole can’t make money; there’s nowhere for them to make money from. And yet they provide a very valuable service! Kalshi likes to point out that “roughly three-quarters of its users have never deposited funds, using the platform to track prices rather than trade them.” But those people, including me, don’t pay for that value. All the zero-sum Kalshi traders are providing useful information as exhaust, and not capturing its value.

What is the right way to capture this value? I would have happily paid Kalshi $20 for the night if it had paywalled its election markets, but I don’t think that would work for its business model. Bloomberg’s Hannah Miller wrote last week that “a number of news organizations, including CNN, CBS News, Dow Jones, and CNBC have struck business partnerships with prediction-market operators like Kalshi Inc. and Polymarket to augment their traditional coverage, with some featuring market trades predicting who’ll win control of the House of Representatives or the hotly contested Texas Senate race,” but she adds that “the cash-flush prediction markets represent a new source of revenue for media companies”: Apparently the prediction markets are paying news organizations for product placement, rather than getting paid for, you know, providing the information that the news organizations’ customers actually want.

The actual answer is that Kalshi and Polymarket try to leverage the goodwill and social utility of their election work to become:

1. Regulatorily favored sports gambling sites for teenagers, and
2. General-purpose financial-derivatives exchanges for trading stocks, commodities, crypto, compute futures and whatever else you’ve got.

The first answer annoys me a lot but has an obvious intuitive appeal. People like gambling on elections, and their election gambling is socially useful. (It tells you who won.) However, they only like gambling on elections every two years, which is kind of a rough business model. People also like gambling on sports, and they do it every day. Yadda yadda yadda. Is the sports gambling socially useful? I posit that (1) everyone knows who won the sports game as soon as the sports game concludes, (2) nobody needs to know who will win the sports game before the sports game concludes and (3) the whole point of sports is that you don’t know who wins until it’s over. Like, that’s why it’s fun? That’s why you watch? The point of sports is that you watch the game to see who wins? That is the point of Election Day for some politics fans, but not, I think, for most people. Most people want to know who won and go to bed. I digress.

Also the sports gambling can be addictive and socially destructive in a way the election gambling can’t be, for various reasons of which surely the main one is that the elections are every two years and the sports are every day.

The second answer — “we are good at predicting elections so we should become a platform for hedging AI compute costs” — is less intuitive but considerably more charming. “We’ve found a new form of financial trading that is super cool once every two years, so let’s expand to other novel forms of financial trading, plus like leveraged stocks and crypto.” Fine.

Just, you know, the graph of my sympathy for prediction markets has sharp spikes every two years. Six months ago I was pretty annoyed that the US financial system is building sports gambling exchange-traded funds, and in six months I’ll probably be pretty annoyed about it again. Today, I get it.

Quant code theft

We talked in 2024 about a lawsuit that Jane Street Group brought against Millennium Management and two former Jane Street options traders. The traders, Jane Street alleged, had taken valuable trade secrets with them when they left for Millennium. The traders, and Millennium, disagreed: In their view, all they took with them was the general knowledge and skills of being good options traders. Millennium’s lawyer “compared the two traders to gold miners who ‘really know how to use a pickax,’” saying that “they can guess and take the risk based on the experience in their head of seeing a discrepancy between a stock and an option, for example, and they are willing to make a bet.”

I wrote that, when you leave a trading firm, there is a continuum of stuff you might take with you. On one end, if your job at your old firm was, like, buying and holding Microsoft stock, and you go and do that at your new firm, your old firm can’t really complain. “Microsoft stock” is not a trade secret. On the other hand, “if your big idea at your old job was a complex software system ingesting millions of data points, and before you leave your old job you email the code to your Gmail account” and use it at your new job, that’s bad. The source code from your old quant job seems like a pretty classic trade secret.

Jane Street and Millennium ended up settling, but before they did they had a very funny court hearing in which Jane Street’s lawyers kept saying “these trade secrets are so secret that we cannot even say the product they traded or the name of the country where the trades happened” and the defense lawyers kept saying “when they were doing these OPTIONS TRADES in INDIA, whoops, silly me.” Some of this was, one assumes, just for comedic effect, but there was also a strategic point. Jane Street’s argument that the information was top secret was bolstered by keeping it secret even in court; the defendants’ argument that it was no big deal was bolstered by treating it cavalierly. After the case was settled, the Indian securities regulator weighed in with its own quite detailed description of the trade, which (1) kind of blew up the secret, (2) kind of supported the defense view, insofar as apparently the trade was “buy index options that are obviously super mispriced relative to their underlying index,” which doesn’t seem all that proprietary, and (3) was more broadly bad for Jane Street, in that arguably the trade was “buy options and manipulate the underlying index.”

Anyway here’s another quant trade secrets case, this one in criminal court:

Richard Ho went on trial Monday on charges that he stole the code from his former employer, Headlands Technologies LLC, to set up his own trading firm. Prosecutors allege Ho then used the code to defraud Tower Research Capital into investing with him.

Ho, a 38-year-old Cambridge University computer science graduate, has pleaded not guilty. He faces as much as 20 years behind bars if convicted of the most serious fraud charge against him. Opening statements in the trial are expected to begin Wednesday.

The case is a rare criminal prosecution over the alleged theft of Wall Street trade secrets and one of the first to focus on the code behind trading algorithms.

It’s not quite clear that he emailed the code to himself, but there are allegations that, for instance, his new code contained the same comments as the Headlands code, which does suggest something other than vaguely remembering his old employer’s approach. “Literally cutting and pasting your employer’s code” is kind of the extreme end of quant trade-secrets theft, which I guess is why it is being prosecuted criminally. But, as in the Jane Street case, the dispute is over how secret the secrets actually were, and some of that plays out in theater about how secret they have to be kept in court:

For Chicago’s Headlands, a firm founded by Citadel alumni, the trial could bring new and unwanted scrutiny of a source code the firm says it spent $1 billion to develop. Prosecutors have taken great pains to continue keeping the code secret despite the public nature of a trial. Many hearings in the case have already taken place in closed courtrooms, and that’s expected to be true for the trial itself.

“In reality, all that is necessary is for a single member of the public (or media) to learn about the trade secrets and then share that knowledge with others, including current or future competitors that would be heavily incentivized to obtain it,” Headlands wrote in an August letter to the judge.

But Ho’s defense has signaled that it will argue that quant code isn’t a trade secret at all because the strategies it drives degrade and have to be continually updated to remain competitive. They plan to call as an expert witness Mark Hendricks, a University of Chicago math professor who has worked with dozens of quant firms.

According to court filings, Hendricks is expected to testify that Headlands’ code has no economic significance because it involves common trading concepts that exist at other firms and are “all known or readily ascertainable by others in the industry.”

Print out the code, scatter it around the courtroom, blow your nose with it. “Oh this code? Man everyone knows this code.” 

Buying stock in college athletes

I dunno man:

LevelUp Sports, founded by college football analyst Bud Elliott, announced plans for potential offerings on Republic that would give individual investors access to securities tied to specified future college-athlete income. Each planned offering is expected to include a security linked to an athlete’s income-sharing agreement with a collectible digital card.

LevelUp provides athletes with capital upfront in exchange for an agreed share of future college income. Covered income may include name, image and likeness (NIL) payments and college revenue-share pay. Professional earnings, cost-of-living payments and W-2 employment income are excluded.

“College athletes are creating significant economic value through NIL, yet the investment opportunity around that value is still in its early stages,” said Andrew Durgee, co-CEO of Republic. “LevelUp is bringing a new model to market that connects investors with the future earning potential of athletes, enabled by the securities infrastructure and tokenization technology of Republic. We’re excited to help bring this new category of sports investing to the market.”

We have talked many times around here about what I sometimes call “the great late-night dorm-room question of financial capitalism,” which is “what if people could sell stock in themselves?” Often, in practice, this is a question about early-career athletes. Many sports have traditionally had a career path in which (1) a lot of very talented people work very hard for essentially no money in the lower levels of the game, (2) some of them then hit it big and make tens or hundreds of millions of dollars as top-level professionals and (3) most don’t. There is an obvious case for a sort of risk-sharing approach:

1. Investors pick like 100 promising athletes and give them $1 million each in exchange for 10% of their future professional earnings.
2. The athletes have money to, like, afford food as they toil in the minor leagues for peanuts.
3. Eventually 10 of them become high-level professionals and make $150 million each, of which $15 million each goes to the investors.
4. The other 90 don’t, and keep their $1 million.
5. Net the investors make $50 million.
6. The 90 athletes who don’t make it are sad about not making it, but at least they got the $1 million. The 10 athletes who do make it are rich, and for the most part don’t begrudge giving the investors a high return on their specific contract. (For the most part.)

This stuff has been rattling around for decades, and in some ways it makes sense that it would be having a moment now. Like, 10 years ago, you would not have had offerings that “give individual investors access to securities tied to specified future college-athlete income,” for at least three reasons:

1. Securities regulation, back then, was kind of strict, and “we will sell shares in individual athletes to retail investors” would have been a real regulatory headache.
2. Kinda seems like weird niche sports gambling, doesn’t it? That was illegal, back then.
3. US college athletes were held to strict amateurism rules that probably would not allow them to strike professional deals on their future income.

But since then the US Supreme Court has struck down limitations on sports gambling and on college-athlete pay, so these deals are more allowable. And US securities regulation is kind of down for whatever these days; selling stock in college athletes to retail investors is nowhere close to the weirdest securities offering I’ve written about in recent weeks.

In another sense, though, it’s a little weird for this particular product to launch now, because … the college athletes … are already getting paid? Like the way it works is that top college athletes can get paid millions of dollars, right now, in NIL payments. They no longer play for peanuts, or for free, in the hopes of one day making it big in professional leagues. They get paid now, sometimes more than they’d make in the professional leagues. If you’re getting $2 million a year in NIL payments for your four-year college career, I suppose you could sell 10% of those payments for an $800,000 up-front payment. But it doesn’t seem especially urgent.

Things happen

AI Data Center Boom Set for Upheaval After US Election Rout. Shares of Elon Musk’s Tesla and SpaceX Tumble After Republicans Are Routed. How audacious London debt trader Hamza Lemssouguer came unstuck. CVS Plans to Break Up $117 Billion Healthcare Conglomerate. “To say AI is going to cure cancer is maybe a little bit of a jump.” Crisis widens at top platinum refiner as probe exposes metal test manipulation. PE Firms Defend Continuation Funds Against Investor Ire. Mark Walter in Talks to Sell Cadillac F1 Stake to Red Sox, Liverpool Owners. Wall Street Banks Are Finally Seeing Their China Bets Pay Off. Beijing Reins In IPO Rush After AI Concerns Hit Stock Market. Sequoia shuts London office as VC firm refocuses on Silicon Valley. Bank Accounts Vanish Just Before Dawn in Shock Paraguay Shutdown. WeRateDogs does not rate AI dogs. AI memecoins. The word of the year is “mogging▸ AI summary.”

Follow Us

You received this message because you are subscribed to Bloomberg’s Money Stuff newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.

|