Programming note: Money Stuff will be off on Monday, back Tuesday.

A fact about human nature that I once did not appreciate, and that I really really really appreciate now, is that it is very fun to gamble on stock options. Like:

1. You can make a lot of money very quickly. It is fun to make a lot of money very quickly, and even if you don’t, it is fun to think about making a lot of money very quickly. You can think about what you’d buy with a lot of money, and that itself can be pleasurable. We talked the other day about “dopamine sites” that let you pretend to order food, because thinking about ordering food is apparently pleasurable even without the food. Thinking about buying mansions, same idea.
2. If you do make a lot of money trading options, you will probably ascribe that to skill. (Correctly or not.) Like, if you win the lottery, you might tell yourself “see it’s because I’m really good at picking random numbers,” but will you really believe that? [1] But if you make a lot of money trading options, it’s probably because you have some sort of system. You are analyzing market data or drawing lines on charts or predicting macro events or corporate earnings or something; you have some reason for the trades you put on. If the trades make a lot of money, you will naturally assume that your reasoning was correct and that you are smart. “I have made a lot of money using my big brain,” you will think, which is probably more satisfying than making a lot of money by luck or by decades of hard steady work. And, again, you can have all of these pleasurable thoughts without actually making a lot of money. “Sure I’ve lost some money trading options, but the next trade will make me a lot of money, and then I’ll really feel smart.”
3. Relatedly, there is something prestigious about making a lot of money trading stock options. A lot of famous billionaires got their start trading stock options. [2]  Some of the most desirable high finance jobs open to new graduates of top universities are in options trading or adjacent fields. So if you are trading options, you are in some sense testing yourself against the best, and if you make a lot of money trading options you can be like “haha, I beat Citadel and Jane Street and Goldman Sachs at their own game, I’m a master of the universe allocating capital around the globe.” And, again, if you don’t do that, you can still imagine what it would be like.

Offset against all of this is the fact that options trading is essentially zero-sum, and negative-sum after transaction costs paid to brokers and market makers. 

And so, if you want to build an efficient machine to extract money from people at scale, selling them stock options is pretty good. There are some obvious design tweaks to maximize the money you can extract:

• Offer frequent options bets. Options that expire each day offer more chances to bet, and more immediate dopamine hits, than monthly options.
• Offer cheap options bets. If people can bet $10 on options, they will, and that will draw them in; if the minimum buy-in is $10,000, you’ll turn away a lot of potential customers.
• Offer the feeling of skill. Sell systems and courses and talk a lot about Greek letters, so that the people trading options get their money’s worth, not in terms of leveraged stock exposure but in terms of feeling smart.
• Offer the feeling of prestige. Emphasize that this isn’t gambling; this is participating in the sophisticated global economy of risk management, blah blah blah, high finance, capital formation, etc.

You see elements of this playbook everywhere. US stock options used to mostly expire monthly, but there has been a huge push to zero-day options (ones that expire in a day), because: more chances to win. And agentic trading, where customers can automatically make hundreds of trades per day, is the next frontier in frequency. US brokerages used to charge commissions for trading stocks or options, but now they don’t, because: lower barriers to entry.

You also see people expand this playbook to other, uh, I guess the polite term is “asset classes,” though my hypothesis here is that the correct term is “bets.” Most notably, a theme of this column over the last year or two has been the huge cultural push, in the US, to make sports betting feel more like stock options. Right now, you can go on your brokerage app and make sports bets, and Robinhood will explain that sports bets are an “emerging asset class,” and US financial regulators will talk about “competitive, fair, and efficient markets that protect the price discovery process of trading in the centralized derivatives markets” for sports bets. And when you bet on sports, the person on the other side of your bets will not be a Las Vegas casino but rather a billionaire’s sophisticated quantitative trading firm that is best known for trading stock options. Your sports bet feels, in many respects, like an options trade, and you can feel smart and prestigious if you win, and imagine feeling smart and prestigious if you lose. (And it’s sports so it’s fun.)

Bloomberg Markets Magazine has a feature on “How India Built (and Broke) the World’s Biggest Options Casino,” which lays out this playbook with great clarity:

“Options-trading websites and apps on laptops and phones were everywhere,” says 27-year-old Deepak Prajapat, who sold women’s suits through Facebook and e-commerce company Meesho before turning to trading. He watched as his hometown transformed from a sleepy dairy hub into something closer to a decentralized trading floor. People traded everywhere — from the doorsteps of mud-brick houses, under the neem trees — three generations often crowded around or hunched over one phone screen. YouTube was the classroom. Dinner conversations, once reserved for crop yields and local wedding gossip, revolved around trading psychology and investor lingo littered with Greek terms such as “theta decay.” …

Admission could cost little more than the price of a flask of chai. … A tiny market move in the right direction could double or triple their money in minutes. A move the other way could wipe out everything they’d put in. …

The NSE and BSE scheduled different index options to expire on different days, giving traders a fresh bet on the market every day of the week. ...

Around this calendar grew a small industry of trading academies. They took over commercial buildings in many cities the same way IT and medical exam-prep centers — previously the touted tickets to social mobility — once did.

This was an extremely successful money-extraction endeavor:

[The Securities and Exchange Board of India, or SEBI] released research in January 2023 — and reiterated it this August — showing 9 out of 10 individual traders in futures and options were losing money, at an average of about 128,000 rupees a year. Across the country, retail traders lost more than 500 billion rupees in a single fiscal year in 2023, with much of those losses flowing to sophisticated firms and professional investors on the other side of the trades.

And then the money extraction got to be too much, so regulators just shut it down:

SEBI cracked down in October and November 2024, roughly tripling the minimum size of an options contract from 5 million rupees to 15 million rupees. ...

That was a meaningful new barrier squeezing out many of the small-ticket traders who’d contributed to the boom. The regulator also limited each exchange to one index with options expiring every week. ... Taken together the changes made options trading more expensive and less frenetic, targeting the ultrashort-term speculation that had fueled the craze.

It worked. Within weeks the most hyperactive derivatives market on the planet had deflated.

The end. There was a large fun casino that extracted a lot of money from retail gamblers, regulators found that bad, they shut down the casino, and it stopped extracting a lot of money from retail gamblers. You could imagine objections:

• “Without small-dollar daily option expiries, retail and institutional traders won’t be able to hedge genuine risks!”
• “Without small-dollar daily option expiries, retail traders won’t be able to build wealth!”
• “Without small-dollar daily option expiries, the Indian equities market will be less complete, and price discovery and liquidity will be worse!”

All of those objections strike me as basically self-refuting but there they are. (“Casinos are fun, these people are adults, and if they want to blow their money on dopamine let them” is a better objection.) Also here’s this:

Prajapat says he doesn’t miss the losses or the days spent staring at flashing prices on his phone. What he misses is the sense of possibility, the idea that geography and circumstance no longer determined who could participate in India’s growing financial markets. For a brief period a smartphone and a few thousand rupees seemed enough to put the people of Gangapur on the same playing field as everyone else.

“It was a beautiful madness while it lasted,” Prajapat says, looking out from his porch. “For a moment we felt connected to the heart of global capitalism, right here in our dusty lanes.”

Right, no, that was the trick! 

I wrote once about egg market manipulation:

There’s a big market (egg producers selling eggs to supermarkets etc.), and there’s a small market (egg producers selling extra eggs to each other on an electronic exchange). The price in the small market determines the price in the big market. Participants in the small market are also participants in the big market. You can spend a little money in the small market to move the price, which can make you a lot of money in the big market.

That’s the core idea of market manipulation: You find some small illiquid market that determines the price of some much larger and more liquid market. You buy $100 million worth of stuff in the large liquid market, without moving the price much. Then you buy $1 million worth of stuff in the small illiquid market, causing the price — in the illiquid market, and also in the liquid one — to double. You sell your stuff in the large liquid market for $200 million (it’s doubled), again without moving the price much. Then you sell your stuff in the illiquid market, maybe driving the price down to zero, but so what: You lose $1 million on your small-market manipulative trades, and make $100 million on your large-market manipulated trades.

In general, this is hard to pull off, because what markets have structures like that? Why would some giant liquid market depend on some small illiquid market for its pricing? There are a few salient examples — eggs, Indian index options, interest-rate swaps in the 2000s — but they’re kind of weird.

But maybe that’s the wrong way to think about it. The other day, I quoted a judge’s opinion in an insider-trading lawsuit, saying:

This is modern trading — where algorithms, AI agents, and career traders are all jockeying, minute by minute, for the newest hot trade, using analyst information, market trends, news reports, scuttlebutt from online forums, and other tea leaves to make split-second decisions.

You could tell a story like this: The marginal prices of stocks are set by, you know, four hedge funds; they are set by “algorithms, AI agents and career traders” reading “tea leaves to make split-second decisions.” The stock market is big and liquid. But the tea leaf market is weird and small. The marginal price setters in stock markets are looking to some data sources to set prices, and those data sources might be small and niche and manipulable, and if you can manipulate them you might have a big impact on stock prices.

Arguably the biggest market in the world right now is, like, “AI.” Trillions of dollars of stock market capitalization, of data-center financing, of expected capital expenditures and revenues, all depend on the path of artificial intelligence adoption. The modern debt market is basically built on the value of computer chips as debt collateral. If there is news suggesting that AI progress will be faster or slower than expected, that causes huge shifts in value.

Meanwhile we’ve talked occasionally about compute futures. Several exchanges are working on developing financial futures products to price and hedge the expected future cost of computing power. This has an obvious use case in the AI buildout; I wrote once:

If you can lock in the future price of computing capacity, then building computing capacity is a less speculative endeavor. You’re not building a data center hoping to sell compute to the AI startups of the future; you’re building a data center knowing that you can sell compute at the futures price that you’ve locked in.

But this is all pretty early and small right now; the compute futures are more of a proof of concept than a robust liquid market for AI computing.

Here is a fascinating story from Semafor reporting that “the US Commerce Department last month ordered Kalshi to take down one of its products tracking the price of AI compute,” and has “also pushed the Commodity Futures Trading Commission … to effectively freeze approval of new compute contracts for 60 days.” (“’This story is false,’ a Commerce spokesman said,” though.) Why? “It’s unclear why Commerce is worried about the nascent market,” but:

One potential reason floated to Semafor by market participants is that compute futures could be manipulated to show a sharp drop in the cost of older chips, which might destabilize AI stocks and debt markets. Some of these markets are thinly traded, which could lead to volatility even without bad actors.

That is: Maybe you can spend a small amount of money to manipulate the market for the value of computer chips as long-term debt collateral. The whole AI economy is, arguably, built on that value. If it goes down, maybe hedge funds will notice. Maybe that will crash the prices of AI stocks and bonds. Maybe you can make a lot of money in the biggest market by spending a little money in the compute futures market.

When a company wants to borrow money, its lenders will want to know if it can pay them back. The main way to figure this out is by comparing the company’s earnings to its debt. The lenders will take some measure of the earnings (often Ebitda, earnings before interest, taxes, depreciation and amortization) and compare it to some measure of the debt (often the total amount of debt, or the interest expense). If the company made $50 this year and wants to borrow $100 for five years, that’s probably fine: It will earn back enough to pay off the debt in two years, and its $50 of earnings is more than enough to cover, like, $8 or $10 or $14 of annual interest on $100 of debt. If the company made $8 this year and wants to borrow $100, that’s harder.

Of course, this isn’t really what the lenders want to know. They don’t want to know how much money the company made this year. They want to know how much money the company will make over the next five years. For many companies, much of the time, this year’s earnings are the best data point for estimating the next five years’ earnings, but that’s not always true. What if they go dow? Lenders will often protect themselves by including maintenance covenants in their loans: If the company makes $50 a year and wants to borrow $100 for five years, the loan might have a covenant saying “if the company’s earnings fall below $20 in a year, it has to pay the money back immediately.” [3]  That way, if the company becomes a lot riskier after taking out the loan, the lenders can demand their money back early, or at least use that demand as leverage to negotiate other protections.

But, again, the thing that the lenders really want to know is how much money the company will make over the next five years, so there is room for negotiation and interpretation. Like:

• If the company made $15 this year and wants to borrow $100, that sounds risky. But if the company signed a big customer on Dec. 30, and expects to make $40 next year, that’s less risky. The expected future earnings, which are what matter, are $40 per year, not $15. Maybe! Maybe the big customer will change its mind in six months. But you have to start somewhere, and maybe $40 is a better starting point for estimation than $15.
• If a covenant requires the company to make at least $20 a year, and it made $15 this year, that’s bad and maybe a default. But if it would have made $40 this year, except that there was a global pandemic that (1) closed the company’s stores for six months and (2) is all better now, then that’s different. The expected future earnings, which are what matter, are $40 per year, not $15. Unless there’s another pandemic. Or some other event. There are always events. But, again, you have to start somewhere.

One way that this all plays out is that companies and lenders use projected earnings, not just historical earnings, in negotiating loans. But “projected” often sounds kind of fake, so another (also kind of fake-sounding!) way that this plays out is with “Ebitda add-backs.” Basically:

• The company measures its earnings this year, as the best starting point for figuring out its creditworthiness; but
• It adds back to those earnings any expenses that shouldn’t be there, because they are not recurring cash costs that actually reduce creditworthiness.

Of course this still leaves tons of room for negotiation and interpretation. The company will tend to think that many of this year’s expenses not indicative of its future earnings power and should be ignored. The lenders will be more conservative. But, you negotiate, and maybe you reach an agreement. If a loan has maintenance covenants, they will probably require the company to maintain a certain level of “Adjusted Ebitda,” which will be defined as earnings before interest, taxes, depreciation, amortization, and any other add-backs that the lenders have agreed to.

During the Covid-19 pandemic, we talked about the “corona clause,” an Ebitda add-back that was added to some loan documents to basically back out the costs of Covid from covenant calculations. I wrote:

This is obviously good! We are rapidly entering a recession due to an exogenous and hopefully temporary pandemic; foreclosing on every loan to every company because they are facing months without income would obviously make things worse; the right thing to do is for lenders to suck it up and be nervous about the credit quality of their borrowers for a while until things get back to normal. The lenders ought to show a little team spirit here, and the corona clause forces them to.

That was written in March of 2020, and my main point was that a series of debt defaults triggered by covenant breaches wouldn’t really help anyone. With hindsight, though, possibly a better point would have been “the effects of Covid-19 on borrowers’ income really will be temporary, so ignoring those effects might more accurately reflect the borrowers’ actual recurring earnings power.” 

Anyway. Take some arbitrary corporate borrower. It has $100 of earnings this year. You want to know how much money it will make over the next five years. A decent starting point, knowing no specific facts about the company, would be $100 per year, grown at some reasonable rate. We are not currently in a big pandemic that shuts down all economic activity, but we are in the early-ish stages of a possibly world-changing boom in artificial intelligence. So … uh … what about that? You could spitball:

• “For an arbitrary company, AI will allow it to fire all of its workers and replace them with cheap chatbots, causing huge savings. So in calculating its future earnings, you should take the current earnings and add back, like, 90% of employee compensation costs.”
• “AI will allow it to invent incredible new products that cure cancer and usher in global abundance, causing huge revenue and margin gains. So in calculating future earnings, you should take the current earnings and like quintuple them.”
• “AI will simply take over its business: Its customers will stop using its product and start using Claude instead. So in calculating future earnings, you should assume they are zero.”

These are all dumb, and in the real world lenders negotiate with actual companies and try to underwrite the course of their actual businesses. Still it is plausible to think that, for many businesses, AI could cause changes to their costs and/or revenues that are (1) quite large, (2) somewhat knowable and (3) not reflected in current earnings. 

Here’s a fun Bloomberg News story about AI Ebitda add-backs:

When a struggling French property services company reported results recently, the numbers included a line item that raised eyebrows in the credit market: €20 million ($23 million) of additional earnings for future savings from a new artificial intelligence strategy.

Emeria SASU boosted its last-twelve-month Ebitda figure by adding “estimated efficiency gains” for the next two years from a program to deploy agentic AI, according to a presentation seen by Bloomberg News. At least ten investors, lawyers and analysts told Bloomberg News that it was the first time they’d seen potential future benefits from AI as an adjustment in company results. …

“We view Ebitda add-backs related to future cost savings related to the use of AI as fully inappropriate to assess real numbers,” said Benjamin Sabahi, head of credit research at Spread Research. “While such initiatives make sense from a margin and strategy standpoint, these have to be quantified once already generated.”

A representative for Emeria, which is backed by Partners Group Holding AG and TA Associates, said the pro forma adjustments in the company’s results “are standard adjustments that are widely used in credit markets and are fully aligned with the methodology agreed with the company’s lenders.”

The AI adjustment is “based on clearly identified and quantified actions, in particular the automated preparation of documents, data, and client responses, as well as automation of tasks in customer service, accounting, and back-office functions,” the statement said. “Management clearly explained to lenders during the last two quarterly update calls where AI will be applied and how the efficiency gains will be generated. The €20 million adjustment represents only a fraction of the overall potential identified.”

I kind of take Sabahi’s point that future cost savings from AI do not actually increase last year’s earnings, but I also take Emeria’s point that it is pretty conventional, in credit markets, to pretend that they do.

I mentioned last week one of my favorite tropes of financial journalism, a profile of a rich hedge fund manager who is not like other rich hedge fund manager, he’s down to earth and not flashy, and his biggest indulgence outside of work is some hilariously on-the-nose rich-guy thing. “When he goes to his house on Nantucket, he only sometimes flies private,” or “his only real indulgence is several yachts.”

Here is a Wall Street Journal profile of Jacob Coxon, the former Anthropic researcher who became famous for quitting because he thinks Anthropic is going to kill everyone. He’s not like other AI doomers, though: 

He didn’t strike friends as especially likely to leave his job in AI for ethical reasons. He wasn’t even a major player in the tightknit community of AI safety advocates who discuss grim scenarios at happy hours across San Francisco.

Except for …

• He was on the UK team at the International Mathematical Olympiad, and “of the six members of Coxon’s team, three ended up working for AI labs — including one who also recently quit Anthropic.” 
• “Coxon after college fell in with a small but influential community of rationalists,” ah.
• “Coxon lived in London at the Newspeak House, a self-described independent residential college,” ah.

Like, before reading this article, I would have guessed that the No. 1 and No. 2 indicators of “most likely to quit Anthropic over safety concerns” are (1) “rationalist” and (2) living in a group house after college. But actually those are No. 2 and No. 3, and being on the UK IMO team pretty much dooms you to quit Anthropic. And he had all three! 

Anyway I don’t think “guy who quit Anthropic over safety concerns who is not like other guys who quit Anthropic over safety concerns” will ever be as big a genre as “hedge fund guy who is not like other hedge fund guys,” but it is a really good one. And “discuss grim scenarios at happy hours” is perfect.

Warsh Defies Trump’s Calls for Rate Cut as War Stokes Inflation. Treasuries Gain as Growing Confidence in Warsh Calms Market. Inside the White House Tussle to Sway Trump on AI. OpenAI Reports New AI Safety Incidents, Sets Disclosure Plan. Inside OpenAI’s Ad Rush. Novo Nordisk Will Use Anthropic’s Claude for Drug Research. AI Fuels 440% Surge in Hackers Using Blockchains to Aid Attacks. Wall Street warns trading boom is losing steam. Exxon Is Nearing Preliminary Deal to Invest in Venezuela’s Oil Fields. Boehly, Walter Sell Chelsea Stakes to Clearlake After Feud. ‘I had a superpower’: investors pile into mind-reading brain implants. DeepMind Offshoot Emulate Closes In on $700 Million Seed Round. HSBC Cuts Back $38,000-Per-Kid School Perk for Some Hong Kong Bankers. “By the end of the night, it’s us and our prospects ripping cigs in the karaoke lounge.”

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!