Money Stuff: AI Agents Will Pay Attention

My little financials dispersion trade [1]  is:

1. Some retail-facing financial businesses make their money from customers who pay too little attention.
2. Some retail-facing financial businesses make their money from customers who pay too much attention.
3. Agentic artificial intelligence will tend to help customers pay more attention, which is bad or good for business, respectively.

We’ve talked about both legs of this a few times recently. Recent worries about “agentic bank runs” reflect the fact that banks make a lot of their money from depositor inattention: When interest rates go up, bank customers mostly don’t move their money, because it would be a pain, so banks can pay below-market interest rates on their deposits. You could have similar worries about life insurance, or annuities, or mortgages: Lots of financial products are built (and priced) based on assumptions about customer inattention. AI agents, the theory goes, have unlimited attention and energy, and will tirelessly move money around to extract the last basis point from banks. (Here is a prescient 2019 law review article by Rory Van Loo on “Digital Market Perfection,” laying out this worry.)

On the other hand, some retail brokerages are absolutely falling over themselves to get customers to adopt AI agents. Why? Because, simplistically, in modern US market structure, retail trades are profitable for brokerages. (Strictly: Retail trades are profitable for market makers who take the other side of those trades, and market makers pay brokerages to take the other side of their customers’ trades.) More retail trades mean more profits. But the customers have day jobs and family commitments and need to sleep, so they can only do so many trades per day. AI agents, the theory goes, have unlimited attention and energy and don’t sleep, so they will do many more trades. They will tirelessly move money around to pay the last basis point to market makers.

Retail brokerages are actually an interesting case because, schematically, they make money in two ways: from trades (receiving payment for order flow from market makers), and from idle cash balances (investing customers’ cash, earning interest, and paying the customers a below-market portion of that interest). Trade revenue is increasing with attention, so agentic AI is probably good news for brokers on that front. Cash-balance revenue is decreasing with attention, so agentic AI is probably bad news for brokers on that front.

I’ve also idly speculated about agentic sports gambling, which would be just pure windfall for sportsbooks and prediction markets.

Today’s Wall Street Journal has articles about both legs of this trade. First: “Lazy Customers Are Great for Banks. Could AI Change that?”

AI assistants may soon run your financial life. That could be a big problem for banks. 

Americans forgo untold sums of money every year on what some might call a laziness tax, for oversights like failing to refinance their mortgage or incurring a late fee after forgetting to pay a bill. Another silent wealth killer: keeping too much cash sitting in checking or savings accounts that pay little or no interest.

Now, artificial intelligence is threatening to put an end to all that.

Second: “Robinhood’s Next Big Bet: AI Bots That Trade While You Sleep.”

Artificial intelligence was front and center at Robinhood’s annual event for its most-active users. Holding court at the George R. Brown Convention Center in Houston, the brokerage’s executives unveiled a slew of new tools and services. Among the most significant: an in-app AI agent that can answer questions about the market, build an investment strategy from scratch and eventually place trades while you sleep, work or otherwise turn away from the screen. 

So-called agentic trading tools will give everyday traders capabilities previously reserved for hedge funds and other Wall Street pros, Chief Executive Vlad Tenev said. 

Robinhood agents can engage in a conversation based on simple written prompts, conduct research and trade automatically. For example, users can ask their agent to help analyze how computing costs might affect certain companies, then open positions in their stocks based on a series of follow-up questions. 

In another use case, traders might deliver a more-specific command to their agents: “Run the wheel strategy,” a popular stock-options trade that aims to generate income by selling cash-secured puts and covered calls. The latter example will be possible with a coming feature called Loops, which will allow Robinhood agents to run a strategy repeatedly, even when users aren’t logged on. 

Robinhood’s launch is part of an industrywide push to embed AI tools in how everyday Americans research, plan and execute their investments. In decades past, mom-and-pop traders had to call their broker and pay a hefty commission on every trade. Now, company executives say, agents will become the driverless cars of Wall Street: enabling investors to trade stocks, options, bitcoin and prediction-market contracts even when they are not paying attention. 

While you sleep! I simply cannot stress enough how much “enabling investors to trade … options, bitcoin and prediction-market contracts even when they are not paying attention” is the Holy Grail of retail finance. Market makers love to trade options, crypto and sports bets with retail customers, because the margins on those trades are big and the customers are largely uninformed and unlikely to create adverse selection. (Even after they “ask their agent to help analyze how computing costs might affect certain companies”!) If they’re not paying attention, even better.

I once wrote that, “for equity market makers, there are few more beautiful phrases in the English language than ‘day trader and volleyball-programs coordinator’”: What you want, in a customer, is someone who trades a lot with no edge. [2]  But “day trader and volleyball-programs coordinator who trades in his sleep” is obviously better!

We talked a few weeks ago about the National Basketball Association’s investigation into the Los Angeles Clippers. Basically:

1. The NBA has rules capping how much a team can pay a player.
2. The NBA does not have rules capping how much a sponsor can pay a basketball player for an endorsement deal.
3. If the owner of a NBA team happens to give an outside company $48 million for “consulting,” and if that outside company happens to give the star player of that NBA team $48 million for “endorsements,” well. Well!

The NBA found that the Clippers paid a lot of money to outside companies that went right back to Kawhi Leonard, allowing the Clippers to, in effect, pay him considerably more than the maximum NBA salary. This, the NBA concluded, was not allowed.

European soccer leagues have somewhat different, but essentially isomorphic, rules:

1. The Union of European Football Associations has rules capping how much a club can spend. Whereas the NBA’s cap is essentially a flat dollar amount, the UEFA cap is essentially a linear function of the club’s revenues. A club that makes $100 by selling tickets and sponsorships and stuff can spend, let’s say, $70 on player and coach salaries and transfer fees. A club that makes $100 by selling tickets and sponsorships and stuff, but has fabulously wealthy owners, might want to spend $300 on salaries and transfer fees. But the rules more or less prohibit that.
2. UEFA does not have rules capping how much a sponsor can pay a soccer club for a sponsorship deal. And sponsorship revenues can be used to pay players.
3. If the fabulously wealthy owner of a soccer club happens to give an outside company £830 million for “consulting,” and if that outside company happens to pay the soccer club £830 million for “sponsorship payments,” well. Well!

The English Premier League announced yesterday that “an independent Commission has found Manchester City FC guilty of all charges related to serious breaches of the Premier League’s financial rules over a nine-season period.” Here is the (heavily redacted) decision, and here’s a good breakdown of the findings in the Athletic.

If you read the (very fun!) NBA report on the Clippers, you have the basic idea of the Manchester City charges. In 2008, Abu Dhabi United Group, an investment firm owned by Sheikh Mansour, acquired Manchester City, then sort of a middling Premier League team. Since then, Manchester City has won the Premier League eight times. It has won the Premier League eight times because it has very good players and coaches, and it has those good players and coaches because it pays them very well, and it can afford to pay them well because, allegedly:

The Club devised a plan to disguise shareholder funding (i.e. equity contributions paid by ADUG) as ‘commercial partner revenue’ (i.e. AD Sponsorship income) – what we have described in Appendix 15 as the Disguised Funding Scheme. By the Disguised Funding Scheme

a) The Club would enter into AD Sponsorship Agreements with AD Sponsors which contained substantial Recorded Sponsorship Fees at well above FMV

b) However

i) The AD Sponsor would not be liable to pay the recorded Sponsorship Fee, and would not do so. The AD Sponsor would be liable to pay (and would in fact pay) only the Base Sum

ii) The Tagged Sum would be payable by (and would in fact be paid by) ADUG …

Use of the Disguised Funding Scheme enabled the Club to represent in its financial statements as commercial income from AD Sponsors a total of £949.94 million in the seasons from 2009/10 to 2017/18. Of that sum

a) Only £119.25 million represented Base Fees for which AD Sponsors were liable and which AD Sponsors in fact paid. …

b) A total of £830.69 million represented Tagged Sums, payable by and paid by ADUG to the Club. The sums making up that total ought to have been recorded in the Club’s annual financial statements as equity contributions from ADUG.

The core situation is that sports teams’ owners are generally limited in how much money they can spend on their sports teams, but sports teams’ commercial sponsors are not subject to the same limits. Sponsors don’t necessarily want to spend all that much money on sports team sponsorships, because there is maybe a diminishing economic return. But sports teams’ owners sometimes do want to spend unlimited and uneconomic amounts of money on their sports teams, because winning trophies is fun and kind of the whole point of owning a sports team. And if the owners in fact have unlimited amounts of money, there are trades to be done with the sponsors.

Incidentally. Most professional sports leagues have rules like this, capping teams’ spending, because if they didn’t (1) teams might spend unsustainably in a zero-sum race to win championships and (2) the teams with the richest and most profligate owners would win. And those things, the theory goes, would be bad. A lot of people seem to find that logic basically compelling. But you don’t have to. From another angle, this is pretty straightforward cartel stuff, restricting competition to save the team owners money. If you’re the owner of a basketball or soccer team, spending caps are obviously good for you, unless maybe you’re the richest owner in your league and care more about winning than money. But if you’re a basketball or soccer player, spending caps mean that you get paid less than your free-market value. For pretty obvious reasons (money, also winning), you might prefer to play for the teams that break the rules.

The business of a middleman is buying low and selling high. If someone wants to sell you a stock or a bond or a commodity or a painting or a Pokémon or a whatever for $100 right now, and someone else wants to buy it from you for $120 right now, good deal for you. 

It’s less good for your customers. Some big markets have transparency rules that make it very difficult to pull this off. For instance, in the US stock market, stock trades are reported publicly more or less instantaneously. [3]  If I see that Nvidia Corp. stock traded at $227.21 a second ago, I might be willing to sell you my Nvidia stock for $227.15, or I might be willing to buy some Nvidia from you for $227.30, but $215 or $245 are right out. 

In other markets, it is easier to make big spreads. The art market is notoriously opaque, for instance. If you buy a painting from someone for $100 million, there is not necessarily any publicly reported information about that sale. If you go to someone else and say “would you like to buy this lovely painting, it is $120 million,” she will have no objective reason to think that’s the wrong price. She might ask you a question like “well what did you pay for it,” but you can stroke your mustache and give her a withering look and mutter “philistine” under your breath and she will be shamed into offering you $130 million. Or you can lie, maybe, though that is not legal advice. The Pokémon market at my kids’ school works like this, too, only worse.

For a while, some traders of some mortgage-backed securities would lie to customers about the prices they had paid for bonds: They would buy a bond at 70 and then go tell a customer “I’m twisting this guy’s arm but I think I can get it for you at 90,” and the customer would pay them 90, and they’d make a huge profit. These bonds were thinly traded and not publicly reported, and this sometimes worked. Several traders were prosecuted for this, and some of them went to jail, because prosecutors thought it was fraud. Judges mostly disagreed, though, and I guess it’s fine? Not legal advice. I always thought that prosecuting them was a second-best regulatory approach. The better regulatory approach would be to require MBS trades to be reported publicly. In the US, many bond trades are required to be reported publicly to the Trace system, and this more or less solves the problem. If you can go see where a bond last traded, and it traded at 70, you won’t pay 90 for it. 

Now, if a bond trades only very rarely, and a trader buys it at 70 and wants to sell it to you at 90, she could always not report it. Buy at 70, don’t report the purchase to the Trace system, sell to you five minutes later at 90, collect a 20-point profit. This seems bad. Probably her employer wouldn’t let her do that; probably it has automatic systems to report trades promptly, and takes those systems seriously and fires people who try to bury their trades. But if she did manage it, there are two advantages to her:

1. You don’t know the right price, so maybe she can get you to pay 90.
2. If she gets caught, is she really going to go to jail? She’s not lying to a customer. She’s just forgetting to file a form. Elon Musk does it!

Anyway here’s a fascinating New York Times story about the US Commodity Futures Trading Commission dropping an enforcement action in the cotton market. Cotton is kind of like bonds, in that there is a reporting obligation for certain big cotton trades. Schematically, a middleman who sells a lot of cotton has to report it, and then the trades get aggregated into a weekly government report, and the weekly government report is used by other market participants to figure out what the right price is. So if you can sell cotton at $120, and you want to buy cotton at $100, it is helpful for you not to report the $120 trade, because if people don't know about it they might sell to you at $100:

By law, the companies must report their sales contracts to the Agriculture Department, which publishes weekly reports on supply and demand that farmers rely on to price their goods and make other decisions. But if the traders delay in reporting their export contracts — as the whistle-blower said was the case with the two firms — it gives a false impression of actual demand and the farmers set their prices lower, making less money.

But this is all kind of vague and aggregated; it’s not like you have to report every trade to a public tape within 15 minutes. Two commodity trading companies allegedly delayed their reporting: They would enter into big sales contracts, not report them, and then go out and buy cotton from farmers at low prices. Other cotton traders perceived that as, basically, fraud, a way to rip off their counterparties:

O.A. Cleveland, a professor emeritus at Mississippi State University and a national cotton expert, said he had told major cotton traders in industry meetings: “You are trying to hide this information that should be public so you can rip off the growers.”

One of the companies settled with the CFTC just before the end of the Biden administration; the other one waited, and the Trump administration dropped its case. The reasoning is not totally clear, but the Times notes:

Brooke Nethercott, the agency’s spokeswoman, said the Biden administration fined industries billions of dollars for minor, administrative offenses.  …

[Former CFTC Acting Chair Caroline] Pham ... objected to charging companies with what she considered possible “paperwork errors,” as she said when she criticized the fine against the Olam Group.

“They were a bit slow in filling out paperwork” is one way to put it, and sounds like no big deal. “They effectively lied to their counterparties about the prices they were getting for cotton” is the other way to put it.

Investment banking runs on an apprenticeship model, but I have always thought that it’s not an especially well-designed apprenticeship. The approximate way that it works is:

1. Junior investment bankers build financial models and format pitchbooks.
2. Senior investment bankers schmooze and advise clients.

You become a senior investment banker by being a good junior investment banker. But, why? There is some overlap of skills: Building financial models all night builds facility at understanding corporate finance and merger tactics and the drivers of companies’ value; inputting senior bankers’ pitchbook-formatting markups builds an understanding of how to persuade clients. Maybe? There’s a lot of non-overlap, too, and junior bankers are often only indirectly and haphazardly trained in schmoozing clients. There are obvious arbitrages. Banks sometimes hire senior bankers with no junior banking experience but lots of schmoozing skills, and they sometimes outsource the models and pitchbooks to satellite offices or artificial intelligence models with no path to promotion. But, mostly, the junior bankers become the senior bankers, and it can be a hard transition.

Some junior investment bankers are good at their jobs and would also be good at the more senior job: They are detail-oriented and indefatigable at building financial models, but they also see the bigger picture and play golf. Other junior investment bankers are good at their jobs and would be bad at the more senior job: They can build a sweet financial model, but they can’t talk to clients. [4]

The third, most interesting category is that some junior investment bankers are bad at their jobs but would be great at the more senior job. [5]  They are personable, charming, good with clients and good at golf; they have a good high-level grasp of their coverage industry and good financial intuition, but they’re not going to sit around all night hunting for a mistake in a model. They are not much use to the senior bankers who rely on them for accurate timely work. But if they were senior bankers, someone else would do the accurate timely work, and they’d schmooze it on over to the clients and win lots of deals.

These people either get fired in disgrace or survive their analyst program by the skin of their teeth and then go on to run the place. The Wall Street Journal has a profile of John Waldron, the apparent successor to David Solomon as chief executive officer of Goldman Sachs Group Inc., [6]  who almost fits into both categories:

He joined Bear Stearns out of college—after being rejected by Goldman.

“I didn’t do well in the financial modeling test,” Waldron said on Inside Blackstone, a podcast from the private-investment firm. 

Waldron has said that starting his career at a smaller firm enabled him to take on bigger responsibilities and be more entrepreneurial. At Bear Stearns, he owned client relationships at a relatively young age, he said on Goldman’s Exchanges podcast in 2019.

Right the guy who fails the financial modeling test but shows up and immediately owns client relationships is going to be the much more successful banker in the long run. But not what Goldman wants in an analyst.

Susquehanna Says It Will Settle With Alleged Insider Traders. Trump Backs Independent AI Audits in Shift of Safety Posture.  FTC Probing OpenAI, Anthropic Over Product Safety Concerns. KKR warns of growing credit market risks from AI borrowing spree. Nvidia turns to insurers to spread the risk of AI build-out. Wall Street’s Hopes for a Blockbuster IPO Season Are Fading. What It Takes to Get a Job at SpaceX. Griffin Commits $3 Billion to Carnegie Mellon in Largest Single Gift Ever. Griffin’s University Plan Turbocharges Miami Transformation. The Gap Between the Rich and the Very, Very Rich Is Getting Wider. Startups Race to Cash In on Gold Rush Sparked by Whey Protein Shortage.

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!