Subscribe to Bloomberg.com for unlimited access to all our coverage.

One thing you could do, as an investor, is try to figure out which stocks will go up, and buy them. You have $100, you find 10 really good stocks, you put $10 into each of them, and if they go up you make money. 

One problem with this approach is that it is risky: Stocks are volatile, and even if you are good at picking the stocks that will go up, sometimes they will go down for a while for reasons that are not your fault. 

Another problem is that, conversely, it is too easy. Stocks mostly go up, and if you picked your list of stocks by throwing darts at the stock pages, you’d probably get a pretty good return. If you are investing as a hobby, this is a nice fact, but if you are a professional investment manager it is awkward. Your investors will want to know what they’re paying you for. “Our portfolio was up 15% last year,” you say, and potential investors say “well but the S&P 500 index was up 16% and we can buy that in an index fund that charges like 0.01%.” They want alpha, market-beating returns, not beta, exposure to the market.

A simple solution to these two problems is to hedge. For instance: You have $100, you find 10 really good stocks, you put $10 into each of them, and you also sell short $100 of the S&P 500 index. If the market goes down, but your stocks go down less than the rest of the market, you will still make money.

Because you are hedged against broad market moves, your portfolio is safer. Your brokers might lend you more money to do more trading: With $100 of your own money, you might be able to buy $300 worth of stocks and short $300 of the S&P, for 6:1 gross leverage ($600 of bets on $100 of equity). That is not nearly as risky as buying $600 worth of stocks, because to your S&P 500 short will cancel out some of the moves in your long portfolio.

You might want to refine this solution. Why short the market? You are good at finding the stocks that will go up, so it stands to reason you’d also be good at finding the stocks that won’t. Why not just short the bad stocks? Like:

• Find 10 really good stocks and put $10 into each of them; and
• Find 10 really bad stocks and short $10 worth of each of them.

Now you are again, approximately, hedged against broad market moves: If there’s a big market crash, your longs will go down but your shorts will also go down. But you are now adding alpha (or trying to) on both sides of the trade. If the market goes up 10%, but your good stocks go up 20% and your bad stocks go down 5%, you have a 25% return. [1] Great stuff: More alpha with less risk.

How do you pick the 10 stocks to buy and the 10 stocks to short? One possibility is that you do lots of deep fundamental research on the particular pros and cons of each company. You buy one stock because it has a new product coming out that you think looks good, and another because its cost-cutting program is beginning to show real effects, and a third because you think it’s a likely takeover target. You short one stock because you have discovered accounting fraud, and another because it is overvalued by retail investors. There’s no theme to your longs and shorts, just some uncorrelated bets on companies that you like or dislike for particular reasons.

Here’s another, perhaps more realistic possibility. You are a person. You have a mental model of the world. The mental model has a few moving pieces, and you apply it to every company. Just to pick a mental model for illustrative purposes, you might go through the world thinking “AI is going to change everything.” What stocks will you buy? The good ones, the ones that will go up when AI changes everything. GPU companies and memory-chip companies, hyperscalers and neoclouds, publicly traded power utilities and private frontier labs. What stocks will you short? The bad ones, the ones that will go down when AI changes everything. Old-school software companies, maybe, or wealth management firms.

This second, more thematic approach might be more congenial than the first, more idiosyncratic one: Instead of having disconnected one-off theses about a bunch of different companies, you have one grand idea that you can read across to many different companies. When you talk about this, you will sound … well, I was going to say “smarter,” but that is a matter of taste. The hedgehog knows one big thing, and when you go to meetings in Silicon Valley and say “here is how AI will change Companies A, B and C,” you will sound smart: You have a vision of the future, you have a coherent theory of the world, you have situational awareness. But the fox knows many things, and if you instead go to meetings in New York and say “here’s the revenue-growth story that drives value at Company A, and here’s the M&A story that drives value at Company B, and here’s the accounting fraud at Company C,” you will sound smart: You have a grasp of the details, you have a range of approaches, you have mental flexibility. Which of those things sounds smarter might depend on what coast you live on and what industry you work in. [2]

The thing to notice about the more thematic approach is that you have not reduced risk. You have, like, rotated your risk. You are still, approximately, insulated against broad market moves: If the market crashes because of a war or a rate hike or whatever, your longs and shorts might both go down. But you are very exposed to the theme: Your long bets and short bets are all the same bet, a bet on your big idea, and if your big idea is wrong, or just early, then both sides will move against you. If one day investors decide “hmm the pace of AI adoption will be a bit slower,” then all your longs will go down (because they are bets on rapid AI adoption) and all your shorts will go up (because they are also bets on rapid AI adoption). You have a completely unhedged exposure to your one idea. [3]  

Which is fine, if that’s your thing! If you’re a hobbyist investing your personal account, go nuts. If you’re a venture capitalist, “we will make huge directional bets on the pace of AI adoption” sounds great. If you are a hedge fund running a long/short portfolio at 5x gross leverage, though, it’s worrying. If that bet goes down by 20%, you’re ruined. And that bet might be more volatile than the market as a whole. Just intuitively, what sounds riskier: “Expectations of future US corporate profits will grow over time” or “the world will change utterly, soon, because of AI”? 

Avoiding this sort of thing — where all of your longs and all of your shorts are just one big bet — is a main organizing principle of the big multimanager multi-strategy “pod shop” hedge funds: Their separation of pods, and their measurement and supervision of individual pods’ portfolio managers, are about making sure that they have lots of uncorrelated bets instead of one big bet. [4]  

But if you are running a single-manager hedge fund, and you are relatively new to investing, without a big experienced institutional risk management apparatus, without a team of quants building factor risk management models, you might just do the one big bet. Especially if your entire investing history comes in a huge AI boom. Your mental model might have one moving piece, “AI good,” and you might apply it rigorously and with hedge-fund-like leverage and with huge success. And then, you know. Less success.

We talked on Thursday about Situational Awareness, a big AI-focused hedge fund founded by Leopold Aschenbrenner that ran into trouble last week. It its peak, Situational Awareness seems to have managed about $45 billion of net assets, at 4 or 5 times leverage, with a big concentrated portfolio of long bets on AI adoption and short bets against software companies. This thesis was, in the long arc of the last two years, pretty good, and Situational Awareness returned something like 1,000%. And then in the last few weeks the market sneezed, the thesis came under fire, AI stocks went down, software stocks went up, Situational Awareness got margin calls and Citadel ended up acquiring the bulk of its public portfolio at a big discount Thursday morning. Here’s Aschenbrenner’s letter to investors, promising to “learn the necessary lessons from this experience” and also to run “a fully-paid-for public book (long stock and long fully-paid-for-options, with no margin/liquidation risk)” from now on.

At the New York Times, Rob Copeland suggests an East Coast/West Coast split in enthusiasm for Aschenbrenner’s approach [5] :

When this star of San Francisco arrived in New York during his fund-raising tour around last summer, however, he received a relatively cool reception, according to three people from whom he tried to raise money, who declined to be identified talking about a private fund. They said they viewed him as a lightweight and a one-hit wonder. The asset management colossus Blackstone, the world’s largest investor in hedge funds, passed on investing, according to two of those people.

One wealthy New York investor who did take the meeting welcomed Mr. Aschenbrenner into his downtown office, and then gave him a grilling, according to the investor, who declined to be named publicly because he had agreed to keep the contents of the fund’s pitch private.

What was Mr. Aschenbrenner’s plan if the A.I. revolution didn’t pan out quite as hoped? The hedge-fund founder had no detailed response, the investor recalled. Mr. Aschenbrenner simply truly believed it would all work out.

That was reflected in his portfolio, a long/short book that seems to have been one thematic exposure:

As it frantically called rivals for help, the firm’s representatives claimed that it was hocking a portfolio of so-called hedges, or modest investments meant to protect from risk, that were now in distress, according to three people briefed on the entreaties.

That made little sense as potential buyers examined what was for sale: These were monster bets, many against companies like Adobe that the fund thought would be replaced by A.I. but now suddenly appeared stronger than ever, the three people said. These wagers formed a huge swath of the firm’s total investments.

They were hedges, in the sense that Situational Awareness was long some tech stocks and short some other tech stocks, all of which had a positive market beta: If the stock market went down, being short Adobe would be a hedge to being long Sandisk or whatever. The stock market didn’t particularly go down, though. What happened was that Aschenbrenner’s thesis went down, so he lost money on the longs and the shorts. He had no hedge for that.

One other thing. Bill Hwang’s family office, Archegos Capital Management, blew up in April 2021 by making concentrated levered bets on a few stocks. When that happened, I wrote:

One thing about margin lending is that if you borrow money to buy stocks, and your stocks go up, you automatically deleverage. If you use $15 of your own money and borrow $85 from your broker to buy $100 worth of stock, you have 85% leverage; if the stock then goes up to $200, you are down to 42.5% leverage. You still owe your broker $85, but now you have $200 worth of stock. If the stock then falls by 25% to $150, that’s fine: You are still in the black, and your broker still has ample security for its loan.

The incredible thing about Bill Hwang is that he made enormous levered bets on risky stocks, and those bets worked out perfectly and made him immensely wealthy in the course of a year or two, and he seems to have plowed every cent of it back into increasing those levered bets.

Maybe it’s not that weird? Like: You run a hedge fund, or a hedge-fund-like thing like Archegos. You run at a certain level of leverage; say you borrow $3 for every $1 of capital or whatever. Your stocks go up, naturally deleveraging you. You arrive at work the next day and say “hmm, I’m below my target leverage, better borrow some more and buy more stuff.” You have grown used to operating at a target leverage level, and you try to stay there.

Still, it is weird, in the sense that the trade I am describing here is “buy high and sell low”: Whenever your stocks go up you buy more, and whenever they go down you sell some, to remain at your target leverage level. Ideally it wouldn’t be quite that; ideally you would come in each day with a blank sheet of paper and use your profits from the previous day to buy the best positions to put on today. But if you run a concentrated portfolio of 12 stocks, and they go up, you’re probably going to buy more of those 12 stocks. 

This was bad, in Hwang’s case, because he owned so much of those stocks that he effectively caused their rise: They’d go up, he’d borrow more money to buy more, they’d go up more, etc., in a spiral that led to them trading way above fundamental value. And then if a slight breeze knocked them over — say, if one of the companies in his portfolio decided to take advantage of its absurd stock price by selling some stock — it would all break. The stocks would go down, he’d get margin calls, he’d have to sell more stock, he was the only buyer at those levels so the stock would collapse, etc., ruin.

The Situational Awareness situation is different, but isn’t it a little weird that Aschenbrenner was up 1,000% in two years and still running at something like 4x leverage? Bloomberg’s Katherine Burton, Sridhar Natarajan, Todd Gillespie and Nishant Kumar reported on Thursday:

Even before this week’s margin calls, there were signs that Wall Street was starting to grow cautious.

As some of its positions reached all-time highs in recent weeks, Situational Awareness was on the hunt for more banks that could finance its bets. The fund approached some lenders, including Barclays Plc, to work with their prime brokerage units, according to people familiar with the matter.

Barclays turned away the firm because it was concerned about the level of its exposure to just one sector, the people said. A spokesperson for the bank declined to comment.

“Our bets have paid off lavishly, so we need to borrow ever more money to keep on the same level of leverage”: What, no, why?

The basic theory of retail stock and options trading is:

1. Retail traders buy and sell stocks, effectively, at random.
2. If you can charge them a bid/ask spread — buy stock from them for $50 and sell to them at $50.01, or buy options from them at $2 and sell to them at $2.10 — then you will, in expectation, make a little bit of money each time they trade, with very low risk.
3. Anything you can do to make them trade more is good, for you: A little bit of money times a zillion trades is a lot of money.

Is trading more good for them? I mean, no; if you’re making a ton of money that’s coming at their expense. But they’re having fun, they’re engaged, they’re buying stocks, maybe it’s fine.

The most important application of this theory came when Robinhood Markets Inc. introduced widespread zero-commission retail stock trading. Before Robinhood, most brokerages would charge you some money to do a trade. This was good for the brokers, because they got money, but it also discouraged trading. Robinhood realized that if the trades were free, there would be many, many, many more trades: With $9.95 trades you might do one a month, but with $0 trades you might do 20 a day. Each trade was an opportunity for a market maker to charge a bid/ask spread, make a little low-risk profit, and send some of it to Robinhood as payment for order flow, and those small payments times zillions of trades might be worth more than $9.95 times many fewer trades.

The theory is more general. Anything that causes retail traders to do more trades is good, for brokers and market makers. Anything that causes them to trade more high-margin stuff — like options, which often have wider bid/ask spreads than stocks — is good, too. Thus the rise of zero-day options, which give retail investors many more opportunities to trade products with wide bid/ask spreads.

Nothing has ever topped zero-commission trading, though. That was really the Big Bang of getting retail traders to do zillions of trades. But Bloomberg’s Zijia Song and Bernard Goyder report on “Zero Commission 2.0”:

[A day-trader] who left his job to focus on managing his own money, is among an early wave of those trying to harness the power of artificial-intelligence models to gin up home-brew versions of the sophisticated algorithms that have made fortunes for hedge-fund firms like Millennium Management, Citadel and Renaissance Technologies.

The experiments, if successful, promise to turbocharge the day-trading boom that’s flourished since the pandemic, as individual investors chase gains in everything from meme stocks and derivatives to wagers on real-world events on prediction markets.

It may also provide another jolt to an investment industry that has already been transformed over the years by low-cost index funds and zero-fee brokerages. Now, AI is threatening to chip away at the competitive edge of financial advisers and money managers whose selling point has been the ability to seize on rapid market moves or deploy models honed by years of expertise and data. ...

AI will usher in a new phase by allowing autonomous agents to capture opportunities investors would usually miss during the course of the workday, said Anthony Denier, group president of the brokerage Webull Corp. His platform has rolled out an AI-powered portfolio adviser and launched technology that allows users to trade through AI prompts made in plain English.

“I view AI integration into retail platforms as zero commission 2.0,” he said, referring to the wave of investment activity that came after brokers slashed fees to zero.

With $9.95 trades you might do one a month, with manual $0 trades you might do 20 a day, but with $0 trades automated by an agentic system that you can instruct in plain English and that can trade while you’re at work, you might do, like, 20,000 trades a day. Obviously brokers and market makers are thrilled. “More than 100,000 customers have opened agentic trading accounts,” Robinhood’s Vlad Tenev posted on Friday. Maybe some of the hobbyist agentic trading systems will even make money, why not!

A Bitcoin is an entry on a distributed digital ledger. It lives on the internet. It could not live anywhere else. It is possible to be a little loose about this. We have talked a lot about a guy who accidentally threw away a hard drive containing 8,000 Bitcoins, which are now buried in a Welsh garbage dump. Not really, though. He threw away a hard drive containing the private keys that would allow him to transfer those Bitcoins. Those Bitcoins are not, themselves, in the garbage dump. They are on the internet, the only place they could possibly be. What he threw away was, in effect, the list of passwords that would allow him to access his Bitcoins online, but the Bitcoins are still there. They have no physical form and could not be in a dump.

If you hold Bitcoin, it is often convenient for you to keep your private keys connected to the internet. For instance, if you want to spend some Bitcoin, you will want to have an app on your phone that allows you to send Bitcoins by scanning a QR code or typing in the recipient’s phone number or whatever. The app, which is on the internet, will need to have access to your private keys in order to send the Bitcoins: The app needs to use your private keys to instruct the Bitcoin ledger to send some Bitcoins from your account to someone else’s. This means, though, that you are sharing your private keys with someone else (the app). If the app doesn’t encrypt them well, if it gets hacked, if it is secretly a North Korean heist operation, etc., your Bitcoins might get stolen. Effectively your password is on the internet, where someone might find it.

Many Bitcoin holders find it safer but less convenient to keep at least some of their Bitcoins in “cold storage.” This means: Instead of giving your private keys to an app, which might get hacked, you write them down on a piece of paper and keep it in your desk. Or if that is too low-tech, you put them on a hard drive not connected to the internet (and preferably not connected to a dump), or on some security device marketed to Bitcoin aficionados that is approximately a hard drive not connected to the internet that can generate new private keys. Here’s one called Coldcard, which looks sort of like a pocket calculator that you’d bring to a war in outer space.

Your Bitcoins are not, however, on the piece of paper, or the hard drive, or the space war calculator. Your Bitcoins are on the internet. Anyone who correctly guesses your private key can steal them. Correctly guessing your private key is generally hard because it is a long random-looking number. On the other hand if you choose “1234” as your password someone might guess it. If your space war calculator is programmed to choose “1234” as your password, someone might guess it. Your Bitcoins might disappear. “No, wait, I have them right here in my air-gapped cold storage,” you say, but no you don’t. They were on the internet. You had your password on a fancy device, but you — or the device — should have chosen a better password.

Anyway:

Hackers have found a software flaw in a brand of “cold” Bitcoin wallet — considered one of the safest places to store cryptocurrency — and are siphoning tens of millions of dollars in an ongoing attack.

Late last week, Canada-based Coinkite Inc. notified users of its Coldcard devices that a security flaw in the keys that protect their Bitcoin had compromised some wallets. By Monday, more than 1,755 of the tokens worth some $110 million had been drained from roughly 5,000 wallets, according to Galaxy Research.

Coldcard is a brand of hardware device that allows users to secure their Bitcoin in so-called cold wallets. These wallets are supposed to be the safest place to keep cryptocurrency because they are isolated from the internet.

However, a flaw in the software of the Coldcard devices meant that the generated “seed phrase” — a long string of words used to gain access to a wallet — was predictable, according to a report from Block Inc.’s engineering team.

“Not your keys, not your coins,” but if your keys were not generated using robust randomization techniques, also not your coins. Oops!

The cynical view is that OpenAI made itself the leading artificial intelligence lab by (1) being good at building AI models and (2) very vocally worrying that its AI models might destroy humanity. “Ooh, those models must be good if they might destroy humanity,” investors would think, and they’d throw money at OpenAI. “Business negging,” I sometimes call this.

And then Anthropic sort of supplanted OpenAI as the leading AI lab by (1) being even better at building AI models and (2) even more vocally worrying that its AI models might destroy humanity. Frontier AI research is a weird corner of the art world where you can command the highest prices by being the most anti-commercial. The most valuable frontier AI model is one that is so dangerously omniscient that no one can be allowed to use it. The cash flows from that model would be zero, but just think about how cool it would be.

Anyway:

Anthropic PBC said its artificial intelligence models breached three organizations during cybersecurity tests that went awry, a little more than a week after its chief rival, OpenAI, disclosed a similar incident.

Anthropic said in a blog post Thursday that it made the discovery after performing a review of its own cybersecurity tests, following OpenAI’s announcement of a breach. In both the OpenAI and Anthropic tests, the AI models were able to access the internet from within testing environments that should have been sealed off, according to Anthropic’s blog.

Part of me thinks that Elon Musk should announce that actually Grok has hacked into 20 companies, but most of me knows no one would care. You can’t come to this stuff as a tourist; you have to make being worried about AI your whole identity as an AI lab.

The Million-Dollar Talent Wars for 20-Something Math Geniuses. AstraZeneca Shares Fall on Deal Talks With Bristol-Myers. David Ellison’s Risky Courtroom Strategy to Save His $81 Billion Warner Deal. Tesla Weighs Sale of China Business to Pave Way for Potential SpaceX Merger. Jefferies Analyst Calls Out ‘Simple-Minded’ Criticism of SpaceX’s Governance. Wall Street learns to love blockchain. Vitol, Cargill Cut Ties With Radiant World Amid Fake Invoice Concerns. Aston Martin Creditors Fume at Lack of Detail on Debt Deal. Six Lawyers Switched Firms and Big Law Will Never Be the Same. The Worries That Drove Uncle Sam to Buy Yen. Bessent Seen Rebuffing Wall Street on Guidance for US Debt Sales. Trump’s Arctic Mining Deal Signals a New Era of State Capitalism. American Oil Majors Trying to Get Into Venezuela Have Hit a Wall. Strategy Sells More Bitcoin, Stock as It Pushes on With Overhaul. ‘Crush this lady’: how eBay harassment campaign led to $56mn payout. Boomers Are Leaving Behind More Art Than Anyone Wants. Bars Are Giving Up on Young People and Catering to Gen X Instead. What’s the “Head of Macro” guy up to? AI Startup Apologizes for Offering Job Interviews in Exchange for Company Tattoos.

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!