Student loans At the Wall Street Journal today, Cezary Podkul has a beautiful story about bonds backed by student loans. The basic issue is that there were a bunch of securitizations of federally guaranteed student loans, and it became apparent that those loans might not be repaid in full by the maturity date of the bonds. That is bad for investors: They had bought the bonds expecting to be paid off by a certain time, and now there will be a delay. But there's not much they can do about it: The delay is due to a federal program allowing for income-based repayment, in which borrowers can cap their repayments at 15% of their discretionary income and stretch out their repayment period. This is not a negotiated modification of the loan terms that the bondholders could object to; it's just part of the structure of federal student loans, and they have to accept it. Also it's not that bad for them, because the loans are federally guaranteed, meaning that they'll be paid off eventually; the federal government will generally pay any remaining balance after 25 years, or when the borrower dies. It's just that now that might happen after the maturity date of the bonds. For the investors, this is a matter of degree: Getting a government-guaranteed payment a year late is worse than getting it on time, but, in a low-interest-rate environment, it's not that much worse. But for bond ratings agencies, it is a big stark difference: If a bond is paid off by its maturity date, that's good, but if it's not, that's a default. If a bond has a high likelihood of defaulting, then it should not have a high credit rating. Many of these bonds—senior tranches of government-guaranteed student loans—had triple-A ratings. If they were likely to default, they should not have had triple-A ratings: Bond-ratings firms like Moody's Corp. and Fitch Ratings follow strict rules. They will downgrade a security if they don't think it will pay off by the due date, even when the underlying loans are guaranteed by the federal government. "Even in the event all principal will eventually be received after the maturity date, this still must be treated as a default, technical or otherwise," Sandro Scenga, a spokesman for Fitch, said in a statement. ... In 2015, Moody's put approximately $37 billion worth of bonds originated by the private lenders on review for potential downgrades as it revamped its methodology to account for slower repayments. There is a trade here. The issuers—student-loan packagers like Navient Corp. and Nelnet Inc.—do not want their bonds downgraded. The investors who own the bonds do not want them downgraded: That might reduce their market value or make it harder for them to hold on to the bonds.[1] And there is an incredibly simple fix: You just extend the maturity date. If the bond matures in 2026, and you aren't sure the underlying loans will be repaid by then, you can just extend the maturity to, say, 2055. (As Navient did for one of its bonds.[2]) You don't have to change anything else about the bond: All the economic terms and payment provisions and everything else can stay exactly the same; the money will still come in to Navient and go out to the investors exactly as it did before the amendment. It's just that if some of that money doesn't come in until 2027, or 2054, then instead of calling that a "default" you would call it "fine." And you thwart Moody's, because if the bonds aren't due until 2055 then it doesn't have to worry about defaults, so it doesn't have to downgrade them. Or, if it already downgraded them, it has to upgrade them again: Some bonds went on a ratings roller-coaster ride, including a $406 million chunk of triple-A debt that Moody's downgraded to junk on Nov. 1, 2016. Later that month, the maturity date was moved from 2026 to 2055. Within weeks, Moody's upgraded the bond back up to triple-A. Nice work everyone. It's worth saying that, around the time of the downgrade, those bonds never traded below 96.5 cents on the dollar, according to Bloomberg data. (Remember, they're credit-enhanced tranches of government-guaranteed loans.) In a sense, the maturity extension was "right" and the downgrade was "wrong"; the market said that these bonds weren't junk, and so the investors and issuer got together on a technical fix so that they weren't rated like junk. One way to read this is as a good story of financial engineering. I use "good" in a pretty loose sense; obviously one aspect of the story is that there are lots of people who have such overwhelming student debt that they will be making payments until they die and still not really make a dent in the balance. But that is just sort of a background fact for the story of the bond maturity extension; the loans would be crippling regardless of what happened with the bond maturity. The point here is that there was some minor event with smooth continuous economic consequences—the bond's repayment schedule got a bit longer so it became worth a bit less—but sharp discontinuous consequences under a quasi-regulatory regime: The bond's repayment schedule got a bit longer so its ratings got dramatically worse. And so the financial engineers got together and found a strange-looking way to just get back to the normal, rational result, to have the quasi-regulatory regime reflect the economic reality. (Another way to read this story is as a reminder of something I say from time to time around here, which is that bond investors generally prefer that their bonds have high credit ratings. A lot of people worry that our current system, in which issuers rather than investors pay for bond ratings, gives ratings agencies an incentive to rate bonds too generously: If they give out lots of triple-A ratings, they'll be popular with issuers and will get more business. My point is that they'll be popular with investors too: Investors would generally prefer to buy risky bonds with triple-A ratings rather than risky bonds with single-B ratings, because the triple-A ones will be easier to hold and finance due to various ratings-based regimes. Everyone kind of wants the grade inflation. Here the grade inflation is probably justified, but I'll note that "bond fund manager TCW Group Inc. reached out to Navient to figure out how to avoid downgrades on the securities": The investors were the ones pushing this ratings-based modification, not the issuers.) There's one other wild thing about this story. It is traditionally very hard to extend a bond's maturity date. It is viewed as the sort of fundamental economic term that can't be changed without the permission of the holder. So unlike other bond provisions that can be modified with the consent of a majority or two-thirds or whatever of the bondholders, changing the maturity of these bonds generally requires the consent of 100% of the holders. Getting 100% of the holders of anything to agree on anything is generally considered impossible; even if it is obviously in their best interests, just finding them and getting them to mail back the consent is challenging. But here it (sometimes) worked! Getting 100% approval from bondholders was tricky since it is hard to know who owns a bond at any given time. Navient reached out to investors at industry conferences and at its own investor day meetings, worked the phones and used a social network to identify owners and ask if they would be willing to extend the final maturities by many years, often decades. … In total, Navient managed to get 100% bondholder approval on 62 securities with about $9.1 billion outstanding, or about 14% of its $65.7 billion book of federally guaranteed student loan bonds. Nelnet won approval for about $2.4 billion, or about 12% of its book, according to data compiled by the companies and Wall Street Journal research. Goldman is a bank Before today, Goldman Sachs Group Inc. reported its financial results in four segments. There were Investment Banking and Investment Management, which were basically what they sound like. There was Institutional Client Services, which was the sales and trading division. And there was a thing called "Investing and Lending," which was the weird and hated stepchild. The Wall Street Journal today describes it as "a roughly $130 billion grab bag of loans and proprietary bets that shareholders never warmed to"; Bloomberg News notes that Goldman has spent years "lamenting that investors discounted profits from that area because they were more difficult to predict." This makes sense if you think of Goldman as an investment bank. (Disclosure: I used to work there, in investment banking.[3]) It has an investment banking division, it has a sales and trading division, it has an investment management division; all of those are traditional divisions of big full-service investment banks, and they generate recurring income of various levels of predictability. But if you are a big full-service investment bank you will sometimes come across opportunities to take big proprietary risks, to invest a lot of your own money in merchant-banking deals and special situations and odd one-off investments. And if you're Goldman you'll take those opportunities, and they will generate lumpy and unpredictable profits, and you'll cordon them off in a special division, and shareholders will be happy about the profits but won't attribute as much value to them as they do to the more predictable divisions. But it's all kind of silly if you think of Goldman as a bank. "Lending"? "Lending" is the weird unpredictable side business that shareholders can't get their heads around? For a bank? Lending is the most basic and straightforward bread-and-butter business for a bank; it's the core of what banks do. Now, when I started at Goldman in 2007, it was not a bank at all. When Goldman starting breaking out "Investing and Lending" in 2011, it was technically a bank holding company, but it was not especially bank-like. It was a big full-service investment bank that happened to own a bank, which allowed it to make loans, but it felt weird about those loans and they went into the weird segment. In 2020 though Goldman is … kind of a bank? Goldman Sachs Group Inc. said it will shuffle the way it breaks down results by division in a bid to highlight growth in its consumer business and get more credit from investors. The firm will report a new segment named Consumer and Wealth Management that will include its Marcus online lending unit and its credit-card venture with Apple Inc. The company will eliminate its investing and lending segment, after years of executives lamenting that investors discounted profits from that area because they were more difficult to predict. The moves, outlined in a company filing Tuesday, will spread the interest income Goldman Sachs receives from its lending efforts across all four of the new segments and make the firm's divisions more comparable to its competitors. The changes may help the bank's effort to show off its areas of growth as a long slump in its biggest business -- trading -- has weighed on shares. Here's the filing. Investment Banking stays, though now it is sort of investment and corporate banking; loans to corporate clients, which were formerly in Investing and Lending, now go in the investment banking segment. Institutional Client Services is now Global Markets; Investment Management is now mostly in Asset Management (which includes equity investments from the old Investing and Lending). And there's a new Consumer and Wealth Management division, which will include the consumer bank and retail asset management. Corporate bank, sales and trading, asset management, consumer bank. That's pretty much how banks go. Goldman is a normal bank now. Chess Scott Alexander at Slate Star Codex has a fascinating post about teaching GPT-2 to play chess. GPT-2 is a language model, an artificial-intelligence tool "with a simple objective: predict the next word, given all of the previous words within some text." It is widely used as a sort of internet toy to write fake texts based on a model; you can put a paragraph of Money Stuff into GPT-2 and get back a new paragraph that kind of follows logically and kind of sounds like Money Stuff. Last year Alexander "argued that the program wasn't just an essay generator, it was also kind a general pattern-recognition program with text-based input and output channels. Figure out how to reduce a problem to text, and you can make it do all kinds of unexpected things." GPT-2 chess is a demonstration of that idea: Last month, I asked [a collaborator] if he thought GPT-2 could play chess. I wondered if he could train it on a corpus of chess games written in standard notation (where, for example, e2e4 means "move the pawn at square e2 to square e4"). There are literally millions of games written up like this. GPT-2 would learn to predict the next string of text, which would correspond to the next move in the chess game. Then you would prompt it with a chessboard up to a certain point, and it would predict how the chess masters who had produced its training data would continue the game – ie make its next move using the same heuristics they would. GPT-2 isn't "playing chess," exactly; it is predicting language. Just like Gmail knows that when you get an email saying "here are the documents you wanted," "thanks!" is an appropriate response, GPT-2-chess knows that when a chess player says "1. e4," "1. … e5" is an appropriate response. It has observed how chess players communicate with each other, and it can predict from context what sorts of communication are expected. It works okay: As far as it knows, it's trying to predict short alphanumeric strings like "e2e4" or "Nb7". Nobody told it this represents a board game. It doesn't even have a concept of 2D space that it could use to understand such a claim. But it still captured my rook! Embarrassing! That's cool, but this is a financial newsletter so you can probably guess where I'm going with this. We once talked about using Gmail's suggested auto-responses as an artificial-intelligence tool to make investment decisions: You type up a list of stocks and email it to yourself, and if Gmail's suggested reply is "sounds great, buy those" then you do. That was mostly a joke. But this is … less of a joke? Like if you trained GPT-2 on a bunch of Money Stuff columns it could probably say some stuff that sounds like Money Stuff. If you trained it on a bunch of Warren Buffett annual letters maybe it would say some stuff that sounds like Warren Buffett? Not just in terms of folksy sex jokes but also in terms of penetrating investment insight? Maybe GPT-2 would digest Buffett's mind, or rather specifically the parts of Buffett's mind that are exposed when he writes prose, and it would use that understanding of his mind to write Buffett-like prose recommending Buffett-like investing decisions? Or if you run an investment firm and you've got a corpus of memos from your analysts recommending investment decisions, why not take the memos that worked out—the ones recommending investments that went up—and feed them into GPT-2? Then have it write you a new memo and see if it's any good? I mean it's still kind of a joke, obviously. The normal approach to artificial intelligence in finance is, you know, the computer looks at stocks that went up and tries to spot patterns, to figure out what the stocks that went up had in common. Doing that through the medium of prose—look at the memos recommending stocks that go up and figure out what they had in common—is a weird form of indirection, a dumb and unnecessary complication. But sort of a charming one? One criticism that you sometimes see of artificial intelligence in finance is that the computer is a black box that picks stocks for reasons its human users can't understand: The computer's reasoning process is opaque, and so you can't be confident that it is picking stocks for good reasons or due to spurious correlations. Making the computer write you an investment memo solves that problem! Do not make corporate parody rap videos! I keep saying this! But people keep doing it! Here is a headline that says "Manulife CEO Gori _________ Executives' Spoof Rap Video," and try to imagine a good verb phrase in that blank. "Manulife CEO Gori Attributes Sales Increase to Executives' Spoof Rap Video." "Manulife CEO Gori Accepts Grammy Award for Executives' Spoof Rap Video." "Manulife CEO Gori Totally Cool With Executives' Spoof Rap Video." All impossible, right? The actual headline is "Manulife CEO Gori Apologizes for Executives' Spoof Rap Video," and that's the best possible way that headline could have gone. The former CEO of Insys Therapeutics is going to prison over executives' spoof rap video! If all you have to do is apologize for your executives' spoof rap video, that is a relatively successful executive spoof rap video! And yet: terrible. Manulife Financial Corp. Chief Executive Officer Roy Gori apologized to employees for a spoof rap video that was part of the executive team's year-end message. The internal video, the third in an annual series featuring different music genres, was retracted after some employees said they were offended by it. The Toronto-based insurer said that the video, in which executives wore hoodies and sunglasses, was meant to be "lighthearted." "We regret creating this video, removed it immediately and held a forum to apologize to and hear directly from our employees," Manulife said in an emailed statement Monday. There are few worse phrases in the English language than "meant to be lighthearted," though "executive team's year-end message" doesn't exactly crackle with joy either. Anyway I'm gonna keep saying it: Do not make corporate parody rap videos! A correction In "Things happen" yesterday I included an item titled "This wearable vest grows a self-sustaining garden watered by your own urine." It was supposed to link to this story, but instead it linked to this item, "A beginner's guide to modern classical music." Honestly I do not especially regret the error. If you clicked on that link you're probably better off with what you got than with what you expected. But a really quite surprising number of readers clicked the link, did not get the urine vest, and emailed me to complain. So, sorry, I guess? I can explain the error—both items were in the same set of assorted links at Marginal Revolution over the break—though, now that I have typed that explanation, I kind of wish I hadn't. It would be more fun as a mystery. Things happen US investors throwing money at hottest start-ups. Why the U.S. Repo Market Blew Up and How to Fix It. How Trump's Trade War Is Making Lobbyists Rich And Slamming Small Businesses. UBS Wealth Unit Revamp Gathers Pace With 500 Jobs Set to Go. Franklin Templeton heads for sixth year of outflows. SEC-fined crypto firm sues law firm for bad advice. Jersey Rail Stations Become Hubs for New York Gamblers. World's Largest Bitcoin Mine Lures New Clients to Texas Hotspot. More Zimbabwe Bank Jobs at Risk as Lenders Push Digitization. Firefighter almost dies from poking at popcorn stuck in his teeth. 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! [1] An investment-grade fund, for instance, might have trouble holding one of these bonds if it got downgraded to junk. Or the article mentions a bank that owned one of the bonds: "A downgrade would mean the bank would have to hold more capital against the bond because it would be considered riskier." [2] The Class A-3 notes of SLM Student Loan Trust 2013-1. Here's the original prospectus, and here are the 8-K and supplemental indenture extending the maturity. [3] Though I did "Derivatives related to Advisory and Underwriting Activity," which are apparently moving to the Global Markets division in the reorganization. | 
Post a Comment