Money Stuff: The Robots Make the Predictions

I wrote last week that, in investing, “There’s no magic, no dark matter, no other source of gains. Everyone’s gains come from (1) economic growth and (2) other people’s losses.” In the aggregate, everyone gets the market return, which comes from allocating capital to economic growth. Some people get more and others get less, but they necessarily cancel each other out. People invest anyway, though, because allocating capital to economic growth is a good long-term proposition.

Prediction markets don’t have that. People put $1 into a prediction market event contract, and at resolution it pays out $1 to the winner. There is no investment in economic growth, no source of long-term returns; everyone’s winnings come from someone else’s losses. 

In the stock market, there are big high-frequency trading firms that make markets for retail investors, selling to investors who want to buy and buying from investors who want to sell. Those firms tend to make a lot of money. As a first cut, the money they make comes from the regular investors. (Who else could it come from?) Retail investors regularly get angry about this. But retail investing is still, you know, good. It is still positive-sum: Retail investors are in aggregate putting their money into companies and profiting from overall economic growth. It is possible, likely even, that (1) big electronic market making firms make billions of dollars off of retail stock investors and (2) those retail stock investors also make billions of dollars themselves. It is nice to play a positive-sum game.

Stock options, I don’t know, man.

Big electronic trading firms also increasingly make markets in prediction markets. What conclusion can you draw from that? The extremely obvious one, surely? “Big electronic trading firms are not idiots or charities, they reasonably plan to make a lot of money in prediction markets, that money has to come at the expense of regular retail traders on prediction markets, and since there is no other source of profits in prediction markets, that means that prediction markets are a negative-sum game for retail traders.” It is embarrassing to type this. “Sports gamblers lose money on average,” yes, duh.

Anyway Bloomberg’s Carolyn Silverman, Nathaniel Popper and Marie Patino report:

Over 100,000 accounts lost at least $1,000 on Polymarket, one of the largest prediction markets, according to a Bloomberg News analysis of every wallet active since the beginning of 2025. That is almost twice the number that made at least that much.

Among the winners, a majority of the profits were raked in by a tiny slice of what look to be automated bots, based on the Polymarket trade records compiled by the data firm Dune. Everyone else, in aggregate, lost $131 million. …

While prediction markets have been described as peer-to-peer, the Polymarket records suggest the role of the sportsbook is now largely being played by the sort of automated, high-frequency traders that have long dominated other financial markets. The most active accounts on the site were a small proportion of wallets, but accounted for most of the trading volume.

Right. Two points here. First, it’s not totally clear to me that the large money-making wallets are all actually traditional market making firms. These days individual hobbyist traders can and do vibe-code automated prediction market trading bots. (See below.) Maybe some of them are good at it? Don’t get your hopes up or anything, but it’s at least possible that the automated market makers in prediction markets are not all well-capitalized professional firms that make markets in traditional financial markets. It is still early days.

Second, I’m being a little bit unfair when I say that prediction markets are necessarily zero-sum (or negative-sum after the market maker’s cut). One important argument for them is that they are hedging markets, insurance, that people buy event contracts to hedge some real-world risk. In that context, losing money on an event contract is fine: It means that your real-world risk didn’t come true, you are better off than you otherwise would be, and the prediction market allowed you to responsibly take a risk you otherwise might not have taken. I am sure that this has happened once or twice on prediction markets, though it is always worth emphasizing that “prediction market” means mostly sports betting.

Never mind then:

Investors in a Blue Owl Capital Inc. fund tendered less than 1% of shares to Boaz Weinstein’s Saba Capital Management and Cox Capital Partners, which had offered to buy them out at a steep discount, according to people familiar with the matter.

The tender offer for shares in Blue Owl Capital Corp. II, one of the firm’s non-traded business development companies, expired at the end of last week with limited take up and wasn’t extended, said the people, who asked not to be named because the information isn’t yet public.

The scant sales to Saba and Cox suggest that the malaise in the $1.8 trillion private credit has its limits, with investors preferring to hold onto their shares rather than cash out now at far less than full value.

We have discussed Weinstein’s tender offer before, including with him on the Money Stuff podcast. Basically Blue Owl Capital Corp. II, a non-traded business development company usually called “OBDC II,” normally offers to buy back up to 5% of its shares each quarter. But as people have started worrying more about private credit, OBDC II got more redemption requests, and rather than meet them it has announced a plan to return all of its capital gradually. Investors who wanted out are stuck for longer than they expected to be. So Weinstein offered to buy them out, at about 65% of net asset value. 

And they mostly said no. There are two possible explanations. One is that his price was too low. On the Money Stuff podcast, Weinstein argued that 65 cents on the dollar was not a particularly conservative price, all things considered. But public BDCs tend to trade at closer to 80% of NAV than 65%, and Blue Owl’s own public BDC — Blue Owl Capital Corp., or OBDC — trades at about 76%. [1]  If the assets are worth 76 cents on the dollar, you can’t get them for 65.

The other possibility is that, structurally, investors in non-traded BDCs really do not want to sell at any discount. Last year, Blue Owl itself tried to get OBDC II’s shareholders to exchange into shares of OBDC, and the shareholders revolted. If they got OBDC shares, they could have sold them freely at any time, but they would immediately realize a 20%+ discount to NAV. If they keep their OBDC II shares they can continue to assume they’re worth 100 cents on the dollar. And then eventually OBDC II will cash them out, over time, at NAV.

The fact that the market-clearing price of private credit funds is about 80 cents on the dollar is of no interest to most holders of non-traded BDCs: They want 100 cents on the dollar. They were sold an investment that is not supposed to fluctuate with market prices. The advantage and the disadvantage of a non-traded BDC is that it never has to trade at the market-clearing price. If you are in a non-traded BDC and looking to get out, this can be annoying: You can’t get all your money out at once. But it can also be nice: Nobody will ever tell you that you can’t get all your money out. You have to wait, but you’ll get all of it. Market fluctuations are of no concern to you. You are pleasantly insulated from volatility.

Anyway yesterday Saba “announced a significant expansion of its investment activities into both public and private BDCs and interval funds,” disclosing stakes in some publicly traded funds and noting that it “is considering providing bids on a number of additional products, including the Cliffwater interval fund and Blue Owl’s OCIC – one of the largest private BDCs in the market.” It apparently doesn’t plan to raise its bids, though; its “strategy targets entry points at discounts of 30-40% or greater to NAV.”

Disclosure: Through a financial adviser, I have a little money in Blue Owl Credit Income Corp. I haven’t tried to redeem or anything. I’d be a little tempted to sell it to Weinstein just to have something to talk about. But we’ll see what he offers.

Elsewhere:

Money managers at Pacific Investment Management Co., Janus Henderson Group Plc and Baird Asset Management say retail clients have been receptive to their simple pitch on the merits of bond funds: solid yields, asset-price transparency and no restrictions on pulling money out.

Right, I mean, that is a different product.

In November 2021, Rivian Automotive Inc. went public at a stock price of $78 per share, valuing the company at about $67 billion. Rivian’s chief executive officer, Robert Scaringe, owned stock and options worth more than $1 billion at the initial public offering price. A few months before the IPO, Rivian’s board of directors also gave him a moonshot pay package that would grant him 20.4 million more options if the stock hit certain price targets: He would earn some shares if the stock hit $110, and he’d get all of the shares if the stock hit $295. At that price, those options would be worth some $5.6 billion, but of course at that price he would have created about $188 billion of value for shareholders, so good deal. [2]

Rivian’s stock actually got to $172 a week after the IPO, but it fell below the IPO price by January 2022 and never got back; none of those options vested. [3]  The stock closed yesterday at $16.72, for a market capitalization of about $20 billion. In fact Rivian has destroyed about $47 billion of shareholder value since the IPO, ah well. 

Scaringe, Rivian’s founder, remains the chief executive officer, but that 2021 moonshot pay package is way underwater. What should Rivian do about it? On the one hand, they were paying for performance, they didn’t get the performance, so they shouldn’t pay. On the other hand, he’s still the CEO. They still want to encourage him to do good work and pursue ambitious goals, though now an ambitious goal would be getting back to the IPO price. The problem of how to compensate a deep-underwater manager is not easy. Going forward, you want to motivate him by giving him targets he can hit: moonshots, but reasonable moonshots. But if the market expectation is “if you don’t hit the targets, we will give you easier targets,” then perhaps he will not be sufficiently motivated to hit the targets. 

And so Bloomberg’s Kara Carlson reports:

Rivian Automotive Inc.’s top executive earned a whopping $402.6 million for 2025 after the electric-vehicle maker approved a moonshot compensation package that could pay out billions of dollars in the coming years.

Chief Executive Officer RJ Scaringe’s compensation included $373.5 million in option awards, more than $26.6 million in stock awards and about $1.12 million in salary, according to a regulatory filing Monday.

The total package likely ranks among the biggest annual CEO payouts ever, reflecting ever-greater compensation for leaders seen as essential to their companies. Scaringe founded Rivian in 2009 and took the company public in 2021, growing it from a scrappy startup into an automaker with national scale.

Rivian in November awarded the executive a compensation package worth as much as $4.6 billion over a decade. …

In connection with last year’s award, the company’s compensation committee canceled a 2021 plan for the CEO, saying it contained goals that were unlikely to be attained, according to a filing.

He didn’t really get paid $402.6 million last year; that number is an artifact of pay-package accounting. The new package is described on pages 26 and 27 of Rivian’s proxy statement. What happened is that the board tore up the 2021 options grant because of “the lack of incentive provided by the 2021 CEO Performance Award due to the unlikeliness of attainment of the associated performance goals,” oops, and gave him a new moonshot grant of 36.5 million options that he will earn if he hits various stock-price, income and cash-flow targets. If he hits all of the targets, those options will be worth about $4.6 billion. The lowest stock-price target is $40, a bit more than half of the IPO price. The highest is $140, a little below where Rivian was trading a week after the IPO. Ah well.

At Bloomberg Opinion, computational hydrologist Darri Eythorsson writes about the automated prediction-market trading system he built in six days using Claude Code. “I described what I wanted in plain language,” he writes. “The AI wrote the code”:

Thousands of people, independently building autonomous trading agents in days, all powered by the same small family of AI models, all deploying across the same markets. This is not hypothetical. It is happening now. …

The actors driving correlated AI behavior in financial markets are software engineers, data scientists, Ph.D. students and, well, hydrologists. We have no reporting obligations. No compliance departments. No capital requirements. No circuit breakers. We are invisible to every monitoring framework. ...

The barrier that once separated retail hobbyists from institutional-grade trading infrastructure has collapsed. The dynamics I am describing scale directly — as will the risks on a global scale. …

The aggregate outcome is a system that nobody designed, nobody governs, and that concentrates correlation risk in a way no previous financial technology has achieved.

We talked a few weeks ago about agentic retail investing: A retail brokerage has rolled out pre-built agentic tools to let its customers automate their trading, and I wrote:

We have talked a few times before about the possibilities for AI to coordinate retail behavior. If retail investors tend to get investment ideas from AI chatbots, I have written, those chatbots might tend to send all the retail investors to the same ideas, so the stocks favored by the chatbots will go up. Similarly, here, if retail investing becomes increasingly agentic, the agents might tend to automate stereotypical retail behaviors. These days, the most stereotypical retail behavior probably is buying the dip, and maybe that is the first one that will be automated. And then if the S&P 500 drops 2% or more in a trading session, every hedge fund will buy at the close, knowing that there’s billions of dollars of agentic retail demand coming the next morning.

This sort of combines those two possibilities: If the stereotypical retail behavior is “I will ask a large language model to automate my trading,” then increasingly the market will reflect the approaches and methods and biases and blind spots of Claude Code.

Which honestly is a little bit cool? I mean, bad, whatever, correlated and predictable and manipulable. But also, like, everyone who runs the big frontier artificial intelligence labs is going around saying that in a few years they will build a superintelligence. What if one good use of superintelligence is deciding how to allocate capital throughout the economy? What if Claude becomes all-wise and all-knowing? What if we just delegated all economic decisions to Claude; what if it wisely decided what projects to fund and what businesses to pursue? How would that happen? How would “we,” collectively, delegate all our economic decisions to Claude? Maybe like this? Maybe all the retail investors will delegate their investing decisions to Claude.

This rhymes a little bit with some stuff I have written about index funds: If all the investors delegate their investing decisions to index providers or Larry Fink, then, you know, the index does all the capital allocation. I wrote a decade ago about a famous Inigo Fraser-Jenkins research note calling indexing “Worse Than Marxism.” In a flight of fancy, I wrote:

The market is the best algorithm ever developed for allocating capital. So far! But it also creates incentives for someone to build a better algorithm. …

[Maybe], in the long run, financial markets will tend toward perfect knowledge, a sort of central planning — by the Best Capital Allocating Robot — that is better than Marxism because it is perfectly informed and ideally rational. And once you have that, you can shut down the market: The game is over, and the Best Capital Allocating Robot won.

Maybe that’s Claude! I mean. I was kidding then, and I’m kidding now. But maybe a little?

Broadly speaking, crypto has become more respectable over the years. In the early days, crypto was for hackers and outlaws and online drug dealers; now it is for banks and retirement funds. Along with the industry, a lot of individual crypto companies — or, you know, projects or decentralized ecosystems or whatever — have gone on their own journeys from scruffiness to respectability.

Still some scruffiness. If you run a crypto project that is on a journey from scruffiness to respectability, but you are still at heart an outlaw, there is an interesting business opportunity. I first proposed it back in 2023, in a somewhat different form, but the basic idea is:

1. You start a scruffy crypto project and raise money from scruffy people in scruffy ways. You do not do rigorous know-your-customer checks; you deal only with anonymous crypto wallets, not verified people. You forbid US investors from putting in money, but winkingly encourage them to use a VPN. 
2. Time passes and you become respectable and successful.
3. Your investors or customers are like “hey great, the project we funded is respectable and successful, now it is time to cash out our profits.”
4. You are like “yes, super, as a respectable and successful company we have all your money, here are our audited financial statements, and our efficient team will give you your money back with your profits.”
5. “Of course,” you add, “as a respectable company, we will need you to fill out certain standard know-your-customer forms and legal certifications. We wouldn’t want to send money to criminals or sanctioned people or otherwise undermine our sterling reputation!”
6. Some of your customers cheerily fill out those forms, and you cheerily give them their money back.
7. Others don’t, for reasons, and you keep it.
8. If you executed Step 1 scruffily enough, you keep a lot of money in Step 7.

I more or less stole this idea from a Guy Ritchie movie. Anyway we have talked about Justin Sun, the crypto entrepreneur, and his feud with World Liberty Financial Inc., a Trump family crypto project. Sun put some money into World Liberty in 2024, when (1) he was under investigation by the US Securities and Exchange Commission and (2) World Liberty was, you know, whatever. Now it is 2026, Sun wants to sell his tokens, and World Liberty won’t let him. The Wall Street Journal reported last week:

Last September … World Liberty said it would allow 20% of their holdings to be traded. Sun’s tokens, however, were blocked from sale. ...

The company told Sun privately that identification information he submitted when he bought the WLFI tokens was inadequate, according to the lawsuit, but World Liberty later refused to provide Sun additional information on the matter.

Here is his complaint, which argues that World Liberty changed the rules after taking his money:

[World Liberty’s] August 22, 2025 X.com post stated for the first time that token holders would have to agree to additional terms imposed by World Liberty to have their tokens unlocked. In a statement about “[e]ligibility,” the post stated that “[o]nly persons that affirmatively make required certifications and accept and agree to the unlock agreement and terms and conditions will be eligible for unlock.”

The agreement referenced in the post was the “WLFI Token Unlock Agreement,” which was initially published on World Liberty’s website on or about August 25, 2025. Among other things, that agreement revealed for the very first time that World Liberty would have the power to selectively freeze users’ tokens. Specifically, the agreement provided in Section 6 that World Liberty may, “in its sole discretion, decline to unlock, restrict access to, or freeze any wallet if it determines that such action is necessary to comply with applicable law, enforce its policies, or protect the integrity of the WLF Protocol.”

These conditions were not included in the TPA, the gold paper, marketing materials, or the governance proposal, and the WLFI Token Unlock Agreement was the first time World Liberty publicly purported to have these powers—none of which Plaintiffs agreed to when purchasing their tokens. In other words, World Liberty collected investors’ money on the promise of giving them tokens and only afterward announced that investors would need to agree to give World Liberty sweeping blacklisting and freezing powers as a condition to unlocking their tokens.

Yeah, I don’t know, doesn’t it sound reasonable to say “we can restrict your trading if such action is necessary to comply with applicable law or protect our integrity”? And yet.

I wrote yesterday:

I will be impressed when a hedge fund uses AI agents to raise money from investors without human involvement. Trading the stocks is the easy part!

Someone sent me “Boardy,” which appears to be agentic AI for introducing people in tech to other people in tech? Which is pretty close? Boardy “just closed an $8M seed round ON HIS OWN,” tweeted its (his?) creator last year? Sure? I do feel like pod shop investor relations must be harder than raising a seed round for an agentic AI for introductions. I feel like there are venture capitalists who get an email like “hello I am an agentic AI” and just send a check immediately without reading the rest.

Polymarket Seeks CFTC Blessing to Bring Main Exchange Back to US. OpenAI Misses Key Revenue, User Targets in High-Stakes Sprint Toward IPO. OpenAI-Linked Stocks Slump on Report It Missed Key Targets. UAE to Leave OPEC in May as Iran War Reshapes Oil Market. Iran Is Flooded With So Much Unsold Oil That It’s Stashing It in Derelict Tanks. Wall Street banks boost Treasury holdings to highest level since 2007. Will Bill Ackman’s Legion of Social-Media Fans Show Up for His IPO? Paramount Requests FCC Approval for Hefty Middle East Ownership. ‘Half-amused, half-ashamed’: the employees facing phone bans at work. Meet the Comedian Poised to Oversee Right-Wing Infowars.

If you'd like to get Money Stuff in handy email form, right in your inbox, please subscribe at this link. Or you can subscribe to Money Stuff and other great Bloomberg newsletters here. Thanks!