Money Stuff

Neural Networks and Suspicious Edge

Also back-office costs, bank capital and giant gold coins.

Neural networks.

I enjoyed this diagram of how hedge funds use neural networks to find trading signals:

It does seem a bit busy, though; I took the liberty of drawing up a simplified version that I hope does not sacrifice too much explanatory power:

The original is from this article about "deep learning" in investing:

Consider a quant searching for factors that might push a stock over a benchmark index. Today an analyst has to manually select factors like price-earnings ratios to test. A quant using deep learning gives the neural network a price target and then feeds the model with raw company and market data. The artificial neurons are math functions that crunch data. As it moves through layers, the neurons self-adjust -- or learn -- to get closer to the goal: finding factors that predict when the stock will hit the target.

There is a famous hacker koan on the subject.

One basic thing to think about artificial intelligence in investing is that there is some finite universe of stuff that you could call "market data," prices and volumes and order-book information and whatever. A lot of energy has always gone into interpreting this data, in the form of human "technical analysis," or of increasingly sophisticated AI/machine-learning/deep-learning/what-have-you computer models that look for patterns. Weak-form market efficiency tells you that, in the long run, you can't make any money doing this, but that doesn't mean that you can't make a nice living while waiting for the long run to arrive. What weak-form efficiency is really telling you is that the computer models will gradually get better and better at mining historical prices for information about future prices, until there's no more useful information to be squeezed out of that data.

Then there's another universe of stuff that you could call "fundamental data." That's stuff like price-earnings ratios: Public companies report their earnings, so you can pull a series of earnings data, pull a series of market data (the company's stock price), feed them into the computer, and see if the computer gets any ideas about how earnings predict prices. Or you can pull every number that every company reports in a financial database, and let the computer figure out which numbers are meaningful and which aren't. All the computers have more or less similar financial databases, so you'd expect them to get really good at mining them for signals, until there is no more useful information to be squeezed out of that data either. 

We are pretty far down both of those paths already. If you start a hedge fund and say "my brilliant idea is to buy stocks that have gone up, because I think they will keep going up," or "I am going to buy stocks that have good earnings, because I think stocks with good earnings should go up," you will sound a bit naive. Those are jobs for a "smart beta" fund these days -- or for a very clever neural network -- not for an intrepid human manager. 

Then there's a much broader universe of stuff that you could call ... everything that happens in the world? That's like: Are people tweeting about a company? What words are in their tweets? Are there cars in the parking lots of a retail company's stores? What color are the cars, and how far from the door do they park? What is the chief executive officer's astrological sign? His golf handicap? What color car does he drive, and how far from the door does he park? One problem with all of this data is that it is hard to know if it will predict future stock prices. That is I suppose where the neural network is useful. But a bigger problem is just recognizing it as data, and then getting it. Your neural network -- or a grad student with a spreadsheet -- can tell you if CEOs' golf handicaps predict future stock prices. But first you have to put the golf handicaps into the neural network -- which means it has to occur to you that the network might be interested. The neural network may be better at asking and answering its own questions about the data than previous machine-learning approaches were. But you still have to give it the right data to question. 


Federal prosecutors have repeatedly, though informally, hinted that Carl Icahn leaked information about his plans to buy Clorox Inc. to his sports-gambler friend Billy Walters. They did it in 2014, in the form of leaks from the FBI to the Wall Street Journal. And they did it yesterday, in a hearing at Walters's insider trading trial. But they have never charged Icahn with anything, and though they have charged Walters with insider trading in other stocks, they haven't charged him with insider trading in Clorox.

Why not? One possibility is that they have no actual evidence that Icahn leaked his Clorox plans. (Icahn, and Walters's lawyer, have both denied any leak.) But there's another, more interesting, possibility, which we discussed back in 2014. Even if Icahn did leak his plans to Walters, it's not at all clear that that would be illegal. This drives people crazy, but: It is perfectly legal to trade with inside knowledge of your own intentions. Not only that: Your intentions are your business, and if you want to tell your buddy about them so he can trade too, you can go right ahead. Trading on material nonpublic information is only illegal if that information was obtained in breach of a duty to someone.

That's not legal advice, and there are complications. Icahn's offer to buy Clorox came through Icahn Enterprises L.P., which has public shareholders, so he may have had a duty to them to keep his plans secret. (Also, if he had planned a tender offer, the rules would be stricter -- but he didn't.) But there's at least a decent chance that if Icahn had called up Walters and said "hey I'm gonna buy Clorox, you should get some," that would be totally legal

Anyway now prosecutors want to introduce evidence of that at Walters's trial:

The prosecutors argued that introducing the Clorox trades would demonstrate a pattern of suspect trading by Mr. Walters.

“We would like the jury to know that getting this sort of an edge, he has done it in other stocks as well,” said Michael Ferrara, one of the prosecutors. “We’d like to say there’s some other suspicious trading as well.”

Walters's lawyer objects, correctly, that even if they had talked about Clorox, "there is no basis to suggest that would violate securities laws."

But of course the prosecutors don't want to discuss the Clorox trades because they violated securities laws. They want to tell the jury about the Clorox trades because they look funny. They are "suspicious." They involve "this sort of an edge." They make it seem like Walters liked to trade when he knew things that no one else knew.

Which you and I know is fine. Trading when you have information that other people don't have is the definition of informed investing. It's what makes people spend millions of dollars on analysts and research and corporate access and neural networks. The purpose of capital markets is to incorporate information into prices, to create incentives for people to find out information and then trade on it. A rule that "informed trading is illegal" -- that edge is illegal -- would cripple the capital markets.

And yet it seems to be the theory that prosecutors believe. Or, in any case, they don't have to believe that having edge is really illegal. (After all, they didn't charge Walters with the Clorox trades.) They just have to believe that a jury will think it's illegal. Which is a pretty good bet.

Bill Gross.

Bill Gross's lawsuit against Pacific Investment Management Co. always struck me as absurd, as a lawsuit, but pretty reasonable as, like, a journaling project. Gross had some issues with Pimco that he worked out by writing them down, and then submitting them to a court as part of a $200 million lawsuit. The story is basically that Pimco was heading toward firing Gross, so he pulled a "you can't fire me, I quit" and walked out the door on his own. He happened to do that just before the end of the third quarter of 2014, so he missed out on his $80 million bonus for the quarter. (For the quarter!) You can see why he felt hard done by: He had earned that $80 million, and Pimco's unjustified decision to fire him deprived him of the bonus he'd earned. That is not much of a legal argument: Pimco hadn't fired him, and the norm in finance is that you're not entitled to a bonus if you quit before it is actually paid. But his feelings were perfectly understandable.  

Yesterday Gross settled his suit with Pimco, for "more than $80 million," which will go to Gross's charitable foundation. So they ... gave him the bonus, then? That makes sense. On the one hand, they probably weren't legally obligated to give him the bonus, since he quit before the quarter was over. On the other hand, it's kind of a nasty move to fire -- or "constructively terminate" -- someone just before his bonus, and then stiff him on the bonus. Of course the breakdown in relations between Gross and Pimco featured a lot of nasty moves on both sides. "PIMCO has always been family to me," said Gross in a statement, "and, like any family, sometimes there are disagreements." Sometimes those disagreements involve tantrums and door-slamming. This settlement lets them patch things up like adults, with everyone uttering kind words through gritted teeth. "PIMCO is dedicating a new 'Founders Room'" at its headquarters to Gross and its other founders. Jon Shazar suggests it should be the bathroom.

Back-office costs. 

A big advantage of the blockchain is the sociological fact that it makes people want to talk about improving back-office processes at banks. As I wrote a while back: "If you announce that you are updating the database software used by a consortium of banks to track derivatives trades, the New York Times will not write an article about it. If you say that you are blockchaining the blockchain software used by a blockchain of blockchains to blockchain blockchain blockchains, the New York Times will blockchain a blockchain about it." So I am excited to inform you that a consortium of banks is working to update the database software that its members use to track trades, and the Wall Street Journal wrote an article about it, and the word "blockchain" does not appear once. I was amazed too:

Talks around this effort are at an early stage but so far have included a number of banks, such as Goldman Sachs Group Inc., Morgan Stanley and Bank of America Corp., the people said. If the idea materializes, it could create a joint venture that allows banks to share trade processes and technology.

The hope is this would be widely used by the industry and eventually trim at least $2 billion from the banks’ annual spending, the people said.

The basic idea is that post-trade processing and settlement processes in the financial industry are a little rickety. Each bank has its own systems, and reconciliation among those systems is often slow and buggy. Getting a bunch of banks together to improve those systems seems like it would cut costs, improve speed and reduce risks. Totally transforming the structure of the financial system by moving all human financial and identity data to a single unified blockchain is also a thing that you could do. But they are mostly separate projects. Just making the databases work better is a fine start.

Is bank capital bad?


Higher capital ratios are unlikely to prevent a financial crisis. This is empirically true both for the entire history of advanced economies between 1870 and 2013 and for the post-WW2 period, and holds both within and between countries. We reach this startling conclusion using newly collected data on the liability side of banks’ balance sheets in 17 countries. A solvency indicator, the capital ratio has no value as a crisis predictor.

That's from a new working paper, "Bank Capital Redux: Solvency, Liquidity, and Crisis," by Òscar Jordà, Björn Richter, Moritz Schularick and Alan M. Taylor. The news isn't all bad; "recoveries from financial crisis recessions are much quicker with higher bank capital." Still, if your one-size-fits-all approach to preventing future financial crises is to require banks to have a lot more capital, this seems like a disappointing result. The problem is that "more bank capital could reflect more risk-taking on the asset side of the balance sheet," because "the more risks the banking sector takes, the more markets and regulators are going to demand banks to hold higher buffers."

There is an intuitive toy story that you could tell here, one that is similar to a lot of other stories in finance. In the 1870s, when the authors' data starts, banks were very well capitalized -- "around 30% capital-to-assets" on average -- but had, in modern terms, no idea what they were doing. They'd just, like, talk to you, and shake your hand, and look over your handwritten account books, and ask the village elders if you were of upstanding character, and if everything checked out they'd give you a loan. They wouldn't even think about buying credit-default swaps against it. There were crises all the time, because bankers took all these dumb risks that they didn't understand. But there was a lot of capital in the system and they muddled out of the crises over and over again. In 2017, there's a lot more science, a lot more ways to track and confine and limit risk, and so crises are rarer. And because they are rarer, banks feel comfortable with a lot less capital. And so the crises, though rarer, are worse.

Elsewhere: "Credit Suisse Group AG said its capital buffers are strong enough to consider alternatives to an initial public offering of the Swiss bank unit, signaling it may be open to selling stock to retain full ownership of its most-profitable business." 

Someone stole a giant gold coin.

Well, they did:

Thieves broke into the German capital's Bode Museum before dawn Monday and made off with a massive 100-kilogram (221-pound) gold coin worth millions of dollars, police said.

Police spokesman Stefen Petersen said thieves apparently entered through a window about 3:30 a.m. Monday, broke into a cabinet where the "Big Maple Leaf" coin was kept, and escaped with it before police arrived.

A ladder was found by nearby railway tracks.

I hope they wore striped shirts and burglar masks and handlebar mustaches, and carried the coin in a big bag with a dollar sign on it, and escaped down those tracks on a hand-cranked rail car accompanied by jazzy piano music. I hope they used the coin to buy gum and get change. What a great crime! There is so much dumb ethereal financial crime these days. Some people use sophisticated computers to hack into global megabanks to steal email addresses. These guys used a ladder to climb into a museum and steal a novelty oversized gold coin with "1 million dollars" literally stamped on its face. (It's worth a lot more than that, but everyone knows it's fun to say "one million dollars.") If you ever find yourself in a position where you are considering committing a financial crime, I want you to ask yourself: Is this financial crime as cool as stealing a 221-pound million-dollar gold coin? If the answer is no, give it a miss. 

People are worried about unicorns.

Here is Maureen Dowd on Elon Musk, who is the founder of Space Exploration Technologies Corp., or SpaceX, the Space Unicorn (Elasmotherium ouranou), and who has even more ambitious plans:

He said that the way to escape human obsolescence, in the end, may be by “having some sort of merger of biological intelligence and machine intelligence.” This Vulcan mind-meld could involve something called a neural lace—an injectable mesh that would literally hardwire your brain to communicate directly with computers. 

The neural lace idea comes, of course, with a startup, called Neuralink Corp. Also: "Sex robots? I think those are quite likely," says Musk.

Me yesterday.

I wrote about the inevitable, somewhat embarrassing fact that the underwriters of Snap Inc.'s initial public offering initiated research coverage with a lot of Buy ratings.

Elsewhere in Snap news: "The lobby group that represents the largest U.K. investment managers is urging FTSE Russell, MSCI Global and S&P Dow Jones Indices to avoid including Snap Inc. in their indexes because it issued only non-voting shares." We have talked about this before: Investment managers might genuinely want to avoid non-voting stocks like Snap's, but have no choice if it is included in the indexes. As the world becomes more indexed, the index providers will increasingly become the main arbiters of corporate governance standards. Which is kind of a weird role for them?

Things happen.

What the Future of S.E.C. Enforcement Holds Under Jay Clayton. An Approach to Financial Instrument Reference Data. Tax on Saudi Aramco Oil Profits Is Cut to Prepare for I.P.O. How Sears CEO Lampert cashes in as stores cash out. Tronc Feud Escalates as Billionaire Investor Demands Access to Records. Unreachable Huishan Executive Exposes China Debt, Bank Risks. FASB Proposal Looks to Trim ‘Hedge Accounting’ Requirements. Once golden, Robert Rubin's hedge fund proteges lose some luster. Deutsche Bank in Bind Over How to Modify $300 Million Trump Debt. Trump to Cancel Obama's Policies Aimed at Paris Climate Pledge. U.S. Travel Industry Fears a ‘Lost Decade’ Under Trump. HFT as an insight into where fintech is going. "Technological change may also provide opportunities for a new kind of personal insurance - for flying cars." The $100,000 Anti-Burnout Program for CEOs. Research: When CEOs Don’t Win Awards, They Make More Acquisitions. Drip-free wine bottles. Batched martinis. Heroin dog. Hockey dog. Alone in the Wild for a Year, TV Contestants Learn Their Show Was Canceled

If you'd like to get Money Stuff in handy email form, right in your inbox, please subscribe at this link. Thanks! 

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

    To contact the author of this story:
    Matt Levine at

    To contact the editor responsible for this story:
    James Greiff at

    Before it's here, it's on the Bloomberg Terminal.