Artificial Intelligence and Agency Bonds
Look, Two Sigma is a big smart hedge fund that has done well and made its founders rich and has prospective recruits wear virtual-reality headsets for some reason. And I am sure they are wonderful at the artificial intelligence and the neural networks and the deep learning and the natural language processing. But this is just a word cloud, no?
At its peak in late 2008, the FOMC devoted nearly 37 percent of its meeting minutes to discussing financial markets, using words such as "securities," "credit," "dollar," "rates," and "mortgage." Since then, the share of the FOMC's attention has trended lower, though it has remained persistently above 20 percent during both the Bernanke and Yellen eras at the Fed. In contrast, the FOMC has reduced the relative fraction of its meetings dedicated to discussing growth.
That's from a Two Sigma paper on "An AI Approach to Fed Watching" (via Business Insider), and I guess the AI approach is to, like, count the financey words in the Fed minutes, and then, I don't know, maybe buy more financey things when there are more financey words. (The paper is mostly about the counting, not so much about what you do with it once you've counted it.) I feel like, if a human counted the appearances of the word "credit" in all the Fed minutes for the last decade, and found a correlation between those appearances and credit assets, and then traded on that correlation, he could make money, but there'd be something less than ... deeply intelligent about it, no? It feels a little like a dumb party trick.
I genuinely don't mean to make fun of Two Sigma, who I am sure are smarter and richer than me (and not giving away their secrets in public papers). But it reminds me of what we talked about yesterday, the fact that the rise in artificial intelligence and machine learning will reduce the explainability of the world. There was a time when smart quantitative hedge funds published long papers full of insights about how the economy worked, because their computers reflected the deep economic understanding of the people who programmed them. Now the smart quant funds just let their computers loose to figure stuff out, and the computers come back with some pattern in the words the Fed uses, and the quants sort of sigh resignedly and say, okay computer, go ahead and trade on that pattern you found, just don't tell us about it.
Elsewhere, "People are more likely to take Uber at 2.1x surge pricing than 2x," so there's really no reason to feel smug about human intelligence. Apparently 2x feels like "they must have seen it was raining and just decided to mess with me," while 2.1x feels like "wow, you know, there must be some smart algorithm in the background." (Those are real quotes from an Uber researcher, by the way.) I feel like my fellow humans need a pep talk. You don't need to have an inferiority complex about algorithms! They are just software! Sure they are better at mining Fed minutes for useful trading information than you are, but just because an algorithm produced something that doesn't mean it's right. Also sometimes algorithms produce round numbers, that is just math.
Still elsewhere: "Watch An AI Learn How To Play 'Super Mario World' By Continuously Failing."
There are three broad categories of price-fixing-and-market-manipulation lawsuits against big banks:
- Cases brought by regulators after lengthy investigations.
- Cases brought by private plaintiffs after the regulators have announced their cases.
- Cases brought by private plaintiffs before the regulators have announced their cases.
Category 1, we talk about a fair amount. One thing that usually happens is that the regulators scour the banks' e-mail and chat archives, find the most flagrant and ungrammatical e-mails, and paste them into the press release in boldface. Little about the manipulation is left to the imagination.
Category 2 is pretty straightforward: Once all the dumb e-mails and chats are public, and the banks are settling with regulators anyway, they don't have much defense against the investors they ripped off by manipulating whatever they manipulated.
But Category 3 is weird, though also kind of boring. Typically these cases are triggered by news reports that regulators are investigating something, and by plaintiffs' lawyers' desire to get in on the action early. These cases tend not to have any direct proof that anything was up: The chats aren't public yet. Instead they rely on some more or less vague statistical evidence: They can't prove that prices were manipulated, but prices look high, anyway, and isn't that kind of suspicious?
Here's a Category 3 case:
Bank of America Corp. and Deutsche Bank AG were among five banks sued over claims that traders conspired to manipulate trading agency bonds issued by government entities and institutions like the World Bank, harming investors who bought and sold the securities.
Here's the complaint. It contains a lot of background on the "supranational, sub-sovereign and agency" bond market, and a lot of background on other things that banks have manipulated recently, just to give the reader a sense of how much banks like to manipulate things. ("Similar wrongdoing in other markets supports the plausibility of Defendants' manipulation of SSA bonds.") And there are some quotes from news reports about governmental investigations. But there's no smoking gun of manipulation. Instead, the plaintiffs hired economists who concluded that the bid-ask spreads on agency bonds were higher than they should have been:
Plaintiff’s experts determined that, given the high level of liquidity for SSA bonds generally, the minimum amount SSA bond traders must charge to make their SSA bond market-making activities worthwhile from a profitability standpoint is between 1.0 and 1.5 basis points. Although an SSA bond trader could charge a higher bid-ask spread, in an otherwise liquid and competitive market, higher spreads could not be sustained without losing business. As a result, under competitive conditions, SSA bond traders would not be able to sustain bid-ask spreads for SSA bonds over 1.0-1.5 basis points. Indeed, this bid-ask spread level would reflect a modest premium compared to bid-ask spreads charged in the sovereign bond segment—e.g., U.S Treasuries or the debt of European national governments—which are also highly liquid instruments.
However, Plaintiff’s experts found through their bottom-up breakeven analysis that the bid-ask spreads for USD and EUR SSA bonds sold in the secondary market were significantly greater than the expected 1.0-1.5 basis point bid-ask spread predicted in Plaintiff’s experts’ models. In fact, USD SSA bonds were quoted at bid-ask spreads of 6 basis points on average. EUR SSA bonds were quoted at similarly high levels. This is a striking variance, despite the absence of significant difference in credit worthiness or liquidity between SSA bond and sovereign bonds.
There is not, at this stage, a lot of detail on the model, but that feels a little thin? The question, though, is whether this case can survive long enough for the plaintiffs' lawyers to get discovery and go rooting around in the banks' chat logs themselves -- or whether it can survive long enough for regulators to announce their own cases and just give the plaintiffs the chats.
The term "hedge fund" encompasses firms ranging from a guy with a dog and a terminal up to institutional thousand-person asset-management behemoths, but it is still funny to me that one of the very most institutional hedge funds -- and one that is getting more institutional by the day -- is not a hedge fund:
Point72 Asset Management, the $11 billion family-office hedge fund led by Steven A. Cohen, has begun targeting college sophomores in its recruiting efforts.
On May 20th, the fund will host its inaugural Sophomore Summit, a one-day investment education program for a highly select group of undergraduate sophomores held at Point72’s Stamford, Connecticut headquarters.
"Point72’s step toward targeting top tier talent earlier on is unprecedented in the hedge fund space," but it is of a piece with the rest of Point72's strategy, which involves a "Point72 Academy," intense recruiting, home-grown talent, lots of lawyers and compliance, and a general institutionalization of its structure. Obviously part of what is going on here may be overcompensation for the other thing that Point72 is known for. (Being shut down, back when it was a hedge fund called SAC Capital, for insider trading.)
But I wonder if there is something else. A lot of hedge funds have gotten out of the managing-outside-money business in recent years, and in many cases they're quite big ones. Soros and BlueCrest and Icahn and others, plus Renaissance Medallion and other funds that have been closed to outside investors for ages. The other day I speculated about "a near future in which running a hedge fund is no longer a way to accumulate dynastic wealth," and in which the big hedge fund managers shut down to run their own money free of investor annoyances and fee pressures, while smaller managers can't grow enough to replace them. In that world, the biggest, most institutional hedge funds -- the ones that most resemble banks or asset managers -- won't be hedge funds; they'll be private family offices running their employees' money.
Also there is a new activist investor lobbying group with a catchy name:
Billionaire investors Paul Singer, Carl Icahn, Barry Rosenstein, William Ackman and Daniel Loeb are setting aside their political differences to launch a Washington-based lobbying group to fight mounting attacks on shareholder activism.
Dubbed the Council for Investor Rights and Corporate Accountability, or Circa, it is the first coordinated effort by activists to make their case to lawmakers and the American public that their investment strategy helps, rather than harms, companies and the U.S. economy.
"It’s the opposite of short term. You need to nurture," says Carl Icahn, about activism.
The New York Stock Exchange had a technical problem yesterday with 199 names traded by KCG as designated market maker, leading other exchanges to trade away from NYSE. This does not seem to have been a big deal. People complain a lot about market fragmentation in U.S. equities, but every so often it is useful to have lots of redundant exchanges.
Elsewhere, "BOX Options Exchange, which operates an electronic trading venue, plans to add open-outcry trading in Chicago later this year," perhaps to attract millennial customers who value authenticity and want their options to be artisanally hand-traded. And millennial finance-video-news thing Cheddar "is literally situated at the heart of the action" on the New York Stock Exchange's trading floor, so probably no one should tell them that NYSE does its trading out of a data center in Mahwah, New Jersey. If I ran NYSE I would totally lease studio space in the Mahwah data center to financial news networks. It'd be a sign of street cred: "Reporting to you live, not from the NYSE's antiquated tourist-trap floor on Wall Street, but from the real heart of the action, here among these nondescript server boxes." I'd be such a good TV producer.
Oh also TradingScreen, a financial software firm, suspended its chief executive officer, Philippe Buhannic, "after an internal probe found 'clear and convincing evidence' that Buhannic 'physically assaulted' a worker in the company’s New York headquarters." Just once I would like the explanation for a NYSE outage to be that a floor trader punched a server.
Here is a fun column from Steven Davidoff Solomon about leaks as a strategy in mergers and acquisitions deals:
Secrecy is supposed to be the sine qua non of deal-making. An early leak can destroy a deal by driving the share price of the target company too high or by alerting other potential bidders.
But sometimes it's the opposite. Like with Yahoo, which the market sort of values at less than zero. Yahoo wants to push its share price too high and alert other bidders! So it is a monsoon of leaks:
In this case, reporting on the process has the effect of amplifying the bidders and the numbers they are putting forward. Take the recent reports that Warren E. Buffett and Dan Gilbert of Quicken Loans are in the Yahoo bidding.
This is the type of news that bankers love. There is such a glow around Mr. Buffett that it will help make Yahoo shine. It shows other bidders that there is a real player with money in the bidding, which will hopefully push up interest and prices.
And here is Ronald Barusch on tax opinions in M&A, which he has always found weird:
I remember asking the head of our tax department why one party to a merger agreement should be able to walk away from a deal when a law firm that’s expected to follow its client’s instructions refuses to bless it. That would seem to give the client significant control over whether it is obligated to move forward with a deal.
And now that concern seems to have come true in the Energy Transfer Equity/Williams Companies merger.
Finally here is Chris Sagers on Staples's and Office Depot's decision not to present any evidence in their antitrust case, which they lost. Nothing here is ever legal advice, but as a fake lawyer I am constantly stunned by real lawyers' decisions not to present their side of the story in court. Why do so few insider-trading defendants testify? Why wouldn't you present antitrust evidence? I know there are reasons but at some visceral level I don't get it.
I guess don't buy a corporate jet while you're being investigated for bribery?
There might be some deeper moral to this story but I am not seeing it:
Asset manager Och-Ziff took out a $49m loan for a corporate jet when it was facing the possibility of a multimillion dollar fine for alleged bribery of Libya’s Gaddafi regime and other African government officials.
Since then Och-Ziff, one of the world’s largest hedge funds by assets, has been forced to borrow a further $120m to prepare for a cash settlement with the US Justice Department that is expected to exceed $200m — a sum more than double its last three years of net profits.
It's trying to sell the jet. Elsewhere: "U.S. prosecutors are examining an alleged transfer of hundreds of thousands of dollars from a former employee of a Malaysian investment fund to Tim Leissner, who was then a senior banker at Goldman Sachs Group Inc.’s Southeast Asia operations, according to people briefed on the matter."
Here is a story about the troubled football career of Robert Griffin III that includes this description of a perfect business meeting:
Griffin called for a meeting. He declined to tell Mike Shanahan what he wanted to discuss, saying only it was important. Griffin, Mike and Kyle Shanahan and quarterbacks coach Matt LaFleur gathered in the offensive meeting room at the team headquarters in Ashburn, Virginia. With the coaches seated, Griffin walked to a blackboard and wrote:
- Change things.
- Change our protections.
- Bottom line.
Griffin instructed the coaches to let him speak uninterrupted and rolled through a list of grievances, stressing that substantive changes had to occur immediately.
Sure, management types will tell you to distribute the meeting agenda in advance, to encourage a free-flowing back-and-forth rather than a monologue, and to keep a positive tone rather than just "roll through a list of grievances." But this way could work too? In any case you cannot argue with that list of headings. If your PowerPoint outline doesn't start with "Change things" and end with "Bottom line" you are doing it all wrong. And it can never hurt to throw in an "Unacceptable."
People are worried about unicorns.
Last October the Wall Street Journal published an article suggesting that the proprietary blood testing technology of Theranos, the Blood Unicorn (Elasmotherium haimatos), might not work. In light of subsequent events it is horrifying and fascinating to go back and read Theranos's response:
We have run more than 3.5 million tests, with patient satisfaction scores from tens of thousands of patients that consistently rate Theranos, on average, over 4.8 out of 5. To our knowledge, we are the first laboratory to publish our prices, lab proficiency-testing scores, customer satisfaction scores, guest visit times, and more on our website.
We are confident in the reliability of our tests, because we comprehensively validate the accuracy of every test we run.
Imagine running a scientific operation and, when your accuracy is questioned, pointing to patient satisfaction scores! That's like a Trump University move. Anyway no those tests were wrong. From the Journal last night:
Theranos Inc. has told federal health regulators that the company voided two years of results from its Edison blood-testing devices, according to a person familiar with the matter.
The Edison machines were touted as revolutionary and were the main basis for the $9 billion valuation attained by the Palo Alto, Calif., company in a funding round in 2014. But Theranos has now told regulators that it threw out all Edison test results from 2014 and 2015.
Whoops! There's at least one anecdote of a person sent to the emergency room over a test result that turned out to be wrong. Theranos even threw out some non-Edison test results obtained "with a traditional machine from Siemens AG that was programmed to the wrong settings by Theranos." It is enough to drive you to drink.
But people are worried about unicorns' drinking habits:
I think it’s time for a broad, honest discussion around alcohol and work. So, you tell me: Do startups have a drinking problem?
I feel like the answer is no; as far as I can tell startups are full of Soylent-drinking Crossfitters. Law firms have a drinking problem. Anyway, though, now is your chance to draw a drunk unicorn and send it to me.
People are worried about bond market liquidity.
One (somewhat exaggerated version of the) bond market liquidity story is that liquidity is so bad that the only way to get a lot of bonds is to buy new issues, either in the deal or in the immediate aftermath, when there is still trading. So new issues tend to pop because everyone wants them. Anyway:
Investors who managed to beat out the competition for a slice of Dell Inc.’s $20 billion bond deal are already being rewarded.
The notes sold by Dell on Tuesday gained more than $300 million by early Wednesday in New York, according to data compiled by Bloomberg. The biggest rise was on Dell’s 30-year bonds, which offered a coupon exceeding 8 percent, a rarity for investment-grade bonds.
But if you own those bonds now, you'll have to hold them for 30 years (because of liquidity). Also there is duration risk.
Bayer Proposes to Acquire Monsanto. FMC Technologies, Technip Agree to $13 Billion Oil-Services Merger. "An unintentional 20th century left-right conspiracy made it all-but-impossible for anyone to take seriously the idea that gains in social justice launched and sustained the era of modern growth, and that enrichment in turn reinforced and promoted further gains in justice, and still does." Happy one-year "anniversary of the Dow Jones Industrial Average’s all-time closing high"! Pay Rule May Give BlackRock a Big Recruiting Boost Over JPMorgan. ‘Blank Checks’ Look for Bargains in Private-Equity Portfolios. Jérôme Kerviel is suing Société Générale. The Tiny Cayman Island Holding $265 Billion in Treasuries. Outlaw Libyan Dollar Traders Thrive While Hiding in Plain Sight. Auto Title Lending: Exploding Toasters. FINRA Fines Raymond James $17 Million for Systemic Anti-Money Laundering Compliance Failures. "We find that top-five executives' ex-ante sale of shares predicts the cross-section of banks returns during the crisis." LendingClub cancels summer intern program as capital flees. A $900 fire-escape tent from Etsy so you can Airbnb your apartment and sleep there, too. Big diamond. Quokkaselfies. Times It Can Be on the Subway Other Than Showtime. MC Hammer is hosting a lip-sync event at Grand Central Terminal today.
If you'd like to get Money Stuff in handy e-mail 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 email@example.com
To contact the editor responsible for this story:
James Greiff at firstname.lastname@example.org