Tricky Bots and Lagging Quants
How to trade.
Here's what you do: You get yourself some recurrent neural networks, and you have them observe how humans negotiate. Then, once your neural networks have carefully studied human behavior, you set them loose to negotiate with each other. This is what some Facebook Inc. and Georgia Tech researchers did, and it went great:
Analysing the performance of our agents, we find evidence of sophisticated negotiation strategies. For example, we find instances of the model feigning interest in a valueless issue, so that it can later ‘compromise’ by conceding it. Deceit is a complex skill that requires hypothesising the other agent’s beliefs, and is learnt relatively late in child development (Talwar and Lee, 2002). Our agents have learnt to deceive without any explicit human design, simply by trying to achieve their goals.
High five, bots! "Facebook teaches bots how to negotiate. They learn to lie instead," says this Wired headline, but why can't it be both? The lesson here might be that an essential element of negotiation -- as practiced by humans and imitated by robots -- is concealing and bluffing about your own objective function, making your counterparty think that you value a thing differently from how you actually value it.
In other news, yesterday a jury mostly acquitted three former Nomura Holdings Inc. traders of fraud for lying to customers about the prices of mortgage bonds, though being mostly acquitted of federal fraud charges is still fairly terrible if you are also convicted a little:
Michael Gramins was convicted of conspiracy and cleared of six fraud counts, while the jury was hung on two other charges. Tyler Peters was acquitted of all nine charges. Jurors cleared a third trader, Ross Shapiro, of eight counts of fraud, but deadlocked on one conspiracy count. Prosecutors must decide whether to retry Shapiro and Gramins on the unresolved counts.
Obviously there is a distinction between what the bots did and what the Nomura traders allegedly did: The bots deceived each other about their own subjective valuations of the stuff they were trading online, while the Nomura traders were accused of deceiving customers about actual prices paid to third parties for those bonds. But it seems like the jury was mostly pretty sympathetic to the traders. And why not? "Deceiving your counterparties about how much something is worth to you isn't fraud," the jury might have concluded; "it's just the essence of negotiation." I mean, bots figured that out by watching humans negotiate; why wouldn't humans draw the same conclusion?
Oh, here's another cool fact about the bot study: "At one point, the researchers write, they had to tweak one of their models because otherwise the bot-to-bot conversation 'led to divergence from human language as the agents developed their own language for negotiating.'" That is also familiar to bond traders! I mean, you can read the chats at issue in the Nomura case; they are not always in precisely what you would call "human language." "Ok don't litvak me," said a customer to Tyler Peters, referring to another bond-trader-lies case and meaning roughly "don't lie to me about the price of those bonds." Like the bots, bond traders have developed their own language for negotiating. Eventually the bots will become clever enough to realize that (1) their counterbots are trying to trick them and (2) they can specify a no-tricking condition. "Ok don't 7r1x0rbot me," one bot will say to another, verbing the name of a particularly notorious trickster bot. The future will be great.
You know who else has taught robots to trade? Everyone, literally everyone, it is the hottest thing going in financial markets; even traditional hedge funds are going quant these days. How's that working out?
Overall, quant funds, which use sophisticated statistical models often developed by Ph.D.s rather than trade based on human research and intuition to find attractive trades, rose 1.44% this year, through May, according to data-tracker HFR. That compares with a gain of 8.7% for the Standard & Poor’s 500 index and a rise of 5.7% for the Vanguard Balanced Index Fund, which invests 60% in stocks and 40% in bonds, highlighting how far quant hedge funds are lagging behind more traditional investments.
Still, "through the first quarter of this year, $4.6 billion of net new money was invested in quant funds, even as over $10 billion was withdrawn from non-quant funds." And the robots keep getting more important:
Systematic strategies have barely budged from near-record participation in U.S. stocks. Meanwhile, fundamental equity long-short managers can’t afford to be anything but picky, considering the market’s narrow leadership.
The result: the largest gap on record between humans’ and computers’ gross exposure to U.S. equities, data compiled by Credit Suisse Group AG show. For now, systematic traders are the dominating force in markets.
There are I think two rough ways to think about computers in markets. One is: Artificial intelligence strategies will continually be adapted to more areas of financial decision-making, so that eventually, like, private-equity buyouts and activist campaigns and distressed-debt workouts will be led by packs of recurrent neural networks. The other is: Computers are really good at systematic statistical-arbitrage strategies and less good at, you know, even fundamental equity long-short. In the first model, "quant" will become increasingly meaningless as a style and performance category: The world will be quant, and different strategies with different performance records and risk exposures will all be run by computers. But the second model is the world we live in for now, anyway, which probably means that near-term quant cycles will be like cycles in any other investing trend: Quant funds will do well, and money will pour into them, and then they'll do poorly.
Elsewhere, here is Josh Brown on the lack of euphoria in the current bull market, which he attributes to basically the fact that robots have taken over the job of being euphoric about bull markets:
There has never been an asset bubble in which the industry that catered to that asset didn’t participate. Wall Street has never had an extended bull market during which everyone spent the entire time worrying.
Can you imagine a real estate boom where the brokers and mortgage people stood on the sidelines, forlorn and only taking part out of obligation? How about a gold boom where the miners told polls every week how bearish they were?
Until now. Job insecurity will do that to people.
The stock market is now 35% passive and 65% terrified.
Speaking of passive investing, we talk from time to time around here about how indexing is not as simple as just investing in the entire global portfolio of financial assets: Even if you decide to invest passively, you are making decisions about what asset classes and geographies to allocate to, and even if you rely on indexes, the index providers are making subjective decisions about what counts as "the market." There are few clearer examples than China:
China’s growing acceptance into international capital markets faces a watershed moment next week with a decision on whether a first batch of stocks listed on its $7tn domestic equity markets will be included into the world’s dominant emerging markets stock index.
If MSCI, an index provider, approves the inclusion of a cohort of Shanghai and Shenzhen-listed A-shares into its main emerging markets index, it will confer an unprecedented recognition upon China’s domestic capital markets and oblige funds all over the world to pour billions into the country’s stocks.
In some simplistic sense, China has a stock market, and it is big, so if you are passively allocating money to investments then you should be allocating some of it to China. But of course there are practical concerns -- about liquidity and regulatory structures and currencies and so forth -- that might give you pause. The important thing is that you don't get to make those decisions, if you are an indexer: You outsource them to an index provider, and then are stuck with what it decides.
Meanwhile, what if you really did just invest in the entire global portfolio of financial assets? You'd make about 4.4 percent a year, according to Ronald Doeswijk, Trevin Lam and Laurens Swinkels:
Using a newly constructed unique dataset, this study is the first to document returns of the market portfolio for a long period and with a high level of detail. Our market portfolio basically contains all assets in which financial investors have invested. We analyze nominal, real, and excess return and risk characteristics of this global multi-asset market portfolio and the asset categories over the period 1960 to 2015. The global market portfolio realizes a compounded real return of 4.38% with a standard deviation of 11.6% from 1960 until 2015. In the inflationary period from 1960 to 1979, the compounded real return of the GMP is 2.27%, while this is 5.57% in the disinflationary period from 1980 to 2015. The reward for the average investor is a compounded return of 3.24%-points above the saver’s.
I am fascinated by this report that "the US Department of Justice is scrutinising whether Barclays breached antitrust laws by promising to stop poaching JPMorgan Chase bankers," because, like the Litvak and Nomura bond-lying cases, it raises deep questions about the interaction between criminal law and social norms. In 2015 Barclays Plc hired Jes Staley, a senior JPMorgan Chase & Co. banker, to be its chief executive officer, and he quickly brought on a bunch of his buddies from JPMorgan to take senior roles at Barclays. But then he stopped doing that, and the Justice Department is investigating "whether Barclays entered into a so-called 'no poach' agreement by promising not to hire more JPMorgan bankers."
The Financial Times reported last year that Jamie Dimon, the head of JPMorgan, called John McFarlane, Barclays’ chairman, to complain about the defections. The FT also said Mr Staley then spoke to Daniel Pinto, the head of JPMorgan’s investment bank.
What would make those calls criminal? If Dimon said "don't hire from us and we won't hire from you," and McFarlane said "you have yourself a deal," then that would be bad. "Going forward, the DOJ intends to proceed criminally against naked wage-fixing or no-poaching agreements," the Justice Department announced last year, and it noted that informal "gentleman's agreements" can still count as criminal conspiracies. ("There have been no improper agreements and we continue to hire from each other," says JPMorgan, and Barclays "insiders vigorously rejected any suggestion that such an agreement was entered into" and pointed out that it keeps hiring from JPMorgan.)
But what if Dimon was like "hey all this hiring is rude and you should stop," and McFarlane was like "I suppose it is rude, we'll stop." What if Dimon called McFarlane and just said "really?" and waited in silence until McFarlane was like "yeah, okay, fine." I have no idea what anyone said to anyone else. But it does seem to me like leaving a senior job at one bank to become the CEO at another, and then poaching a ton of senior people from your old bank all at once, is kind of rude and disruptive to your old friends and colleagues, and you might not want to do it, not because of an illegal "gentleman's agreement" but out of a general social sense of gentlemanliness. (Or you might forget your manners, and be reminded, and repent.) What if Barclays really did slow down its hiring of JPMorgan employees because it concluded that there was a social norm against all that hiring? Isn't a "social norm" just another name for a "gentleman's agreement"? Isn't it a shared informal understanding about what should and shouldn't be done? Does that make it an antitrust violation?
This forfeiture complaint from the 1Malaysia Development Bhd. case is really a perfect shopping list for the modern world; according to the Justice Department, the stuff bought with illicit proceeds of 1MDB-related fraud includes an "8.88-carat fancy intense pink diamond pendant surrounded by 11-carat fancy intense pink diamonds," a bunch of other diamonds, stock in Palantir Technologies and Flywheel Sports Inc., a New York apartment, a movie poster, two movies (rights to "Dumb and Dumber To" and "Daddy's Home"), a Picasso, a Basquiat collage, an Arbus photograph, and a yacht named M/Y Equanimity. I spend a lot of time around here criticizing the naming and nicknaming choices of people caught up in fraud investigations, so it is only fair to say: Whoever bought a yacht named "The Equanimity" with money allegedly stolen from a Malaysian sovereign wealth fund is doing it exactly right. That is how you name your supervillain yacht.
On the other hand ... "Dumb and Dumber To"? This is just a pleasing section heading in a legal brief:
I was a classics major in college, but I cannot quite agree with this piece arguing that hedge funds shouldn't name themselves after characters from Greek and Roman mythology because those characters did bad things:
Fund managers would benefit greatly if they actually read the works they are mining for highbrow names, lest they unwittingly select one from the classical corpus with unfortunate associations. For example Hermes, the god of commerce, luck, and wealth, seems a fitting eponym for an investment fund, as evidenced by the many that bear his name… until one recalls that Hermes was also the patron of thieves and swindlers.
Nope! Fund managers wouldn't benefit greatly if they were deeply familiar with classical literature, because it's not like their customers are either. The extent of most investors' classical-allusion-due-diligence process is pretty much "Hermes, sounds Greek, here's a check." Bulletproofing the classical references in your fund name is more or less impossible -- the Greek gods were shady characters! -- but also a waste of time; classicists probably aren't rich enough to invest with you anyway. Just go ahead and name your fund Priapus Capital and don't worry about it.
Blockchain blockchain blockchain.
This might be my favorite blockchain blockchain blockchain story yet, about a Morgan Stanely research note praising Bank of New York Mellon Corp. for its blockchain efforts:
Morgan Stanley says the custody giant has created a parallel infrastructure using blockchain technology. BDS360 (short for Broker Dealer Services 360) monitors the custodial bank's ledger simultaneously and creates a second, redundant ledger that serves as a backup record. It's been up and running since March 2016.
So: Bank of New York Mellon is a giant trusted counterparty that keeps a centralized ledger of who owns which securities. It also keeps a backup ledger of who owns which securities. That backup ledger is a "blockchain," but it is centralized (BNY Mellon is in charge of it) and private ("All that's left is to roll out is client-facing portions"). And that's a cutting-edge example of blockchain adoption by financial institutions. The blockchain: Might be a useful way for giant incumbent centralized institutions to back up their data!
People are worried about unicorns.
Well, there is "Rape Victim Sues Uber Over Handling of Medical Records," "Travis Kalanick is not ethical enough to steer Uber," "Uber Gaffe Is Latest Drama in Storied Career of Billionaire Bonderman," that sort of thing.
In non-Uber unicorn news, "Slack Technologies Inc. has received recent inquiries about a potential takeover from technology companies including Amazon.com Inc." Slack is a favorite around here because of the meta-unicorn, unicorn-of-unicorns approach that it takes to fundraising: It raises more money than it needs, because there is just so much money available, and then uses some of it to invest in other, less favored startups. "This is the best time to raise money ever," said Slack's CEO in 2015, so he did. I guess if Slack sells, that might be a sign that the party is coming to an end.
Well here's a food trend for you, I guess. It is called "breatharianism":
“Humans can easily be without food — as long as they are connected to the energy that exists in all things and through breathing.”
“For three years, Akahi and I didn’t eat anything at all and now we only eat occasionally like if we’re in a social situation or if I simply want to taste a fruit.”
“With my first child, I practiced a Breatharian pregnancy. Hunger was a foreign sensation to me, so I fully lived on light and ate nothing.”
It's from a New York Post article, and I cannot quite tell at what level it is a hoax.
Amazon to Acquire Whole Foods in $13.7 Billion Bet on Groceries. "Don’t forget your buddies at Goldman Sachs!" Bitcoin Tumbles Most in More Than Two Years After Record Run. BlackRock, Elliott and Pimco Want These Changes to Finance Rules. Criticism Mounts on Madoff Compensation Fund. Investors Play the Risky Role of Lender. If Trump Unlocks $2 Trillion at Banks, Here’s Who May Get It. "U.S. bank lending growth has slipped behind that of the eurozone, as the economic recovery in Europe continues to pick up pace." Google faces big fine in first EU case against search practices. JPMorgan Chase Is Biased Against Dads, ACLU Complaint Says. Snap Chairman Joins Zuckerberg in Strengthening IEX's Tech Ties. Goldman-Backed High-Speed Trading Network Sold for $39.5 Million. U.S. Hedge Funds Walk Tightrope for Sole Control of Co-Op Bank. Former Managing Director At Swiss Bank Pleads Guilty To Money Laundering Charge In Connection With Soccer Bribery Scheme. "It’s not just about the inflectional marking of relative and interrogative pronouns any more, people; it’s about getting more sex!" "Little did she know she was about to be attacked by a rabid raccoon she would end up killing with her bare hands." Digital Hero: This Ethical Hacker Only Orchestrates Cyber Attacks If They’re Justified By The Libertarian Manifesto He Wrote When He Was 19.
If you'd like to get Money Stuff in handy email form, right in your inbox, please subscribe at this link. Thanks!
To contact the author of this story:
Matt Levine at firstname.lastname@example.org
To contact the editor responsible for this story:
James Greiff at email@example.com