Instead of manually identifying an alternative source of securities to satisfy these pended requests, the Demand Team processed approximately 98 percent of the pended requests by relying on a function of Goldman’s order management system referred to as “fill from autolocate,” which was activated by the “F3” key. This function enabled the Demand Team to cause Goldman’s automated model to fill locate requests based on the amount of inventory that existed at the start of the day (i.e., the start-of-day inventory level before any locates were granted), even though Goldman’s automated model had already treated the start-of-day inventory as depleted.
I picture the F3 key as sort of a "make it stop" key. The computer already thought those shares were used up. That's why it sent the request to the humans. But the poor harried humans couldn't deal with the requests. So they just hit F3 to say to the computer, in effect, never mind that you used up those shares, just use them again.
This sounds both utterly understandable -- would you want to process 20,000 stock borrow requests a day? -- and obviously wrong: Surely if you had 10,000 shares of borrow at the start of the day, and the computer loaned out 10,000 shares, and then sent you a request to borrow 1,000 more shares, you don't have a reasonable basis for thinking that you can borrow those extra 1,000 shares. Quite the reverse: The computer already used up the shares, so where will you get more from?
And that is the SEC's theory: The Demand Team had no reasonable basis to keep hitting F3. But to be fair to the Demand Team, it's not as dumb as it sounds. The computer model reduced its inventory for every locate it granted, "regardless of whether the client actually used the locate," and many clients didn't. So when the Demand Team kept hitting F3, "they relied on their general belief that Goldman’s automated model was conservative and that the provision of additional locates would not result in failures to deliver the securities if and when due for settlement."
The SEC finds this general belief unreasonable, but it does seem to have been right: As the SEC concedes, "the Demand Team’s belief that the model was conservative was based on their familiarity with their client’s low utilization rates and that Goldman’s rate of failures remained low and did not substantially change between November 2008 and mid-2013." That is not particularly indicative of a vast naked-short-selling conspiracy.
But it's pretty dumb, so eventually Goldman cut it out. The solution that Goldman ultimately came up with is that it "implemented a new automated model with logic that substantially reduced the number of locate requests pended for Demand Team review." The humans were pretty sure that the computer was being too conservative, so they re-programmed the computer to make it less conservative, and to take up less of their time. The computer just started hitting F3 itself.
This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
The disclosure that I used to work at Goldman Sachs seems relevant here. So: I used to work at Goldman Sachs. Goldman has said in its annual reports that it "was not" (or "has not been") "a significant originator of residential mortgage loans." Incidentally, as of its last 10-Q, Goldman did own about $13.2 billion of "loans and securities backed by residential real estate," which I assume is more about mortgage-backed securities than it is about on-balance-sheet whole loans to rich people. I guess I will throw in an "allegedly" here, though this has been so extensively litigated -- and so accepted in popular culture -- that I don't think anyone will disagree too vocally. In 2010, the Securities and Exchange Commission put out a press release with the headline "Goldman Sachs to Pay Record $550 Million to Settle SEC Charges Related to Subprime Mortgage CDO." That was real money back then! Thursday's agreement in principle is nine times that size, and very far from being a record. Not for doing it again, I mean. It's just, the banks that did bad mortgage stuff get fined multiple times for it. Bank of America's does mention origination though: "violations in connection with the packaging, origination, marketing, sale, structuring, arrangement, and issuance of RMBS and CDOs." Necessarily. See footnote 2, though: It does own some RMBS, and so conceivably some writedowns could come out of its pocket. Or it may not be; maybe the cost of the consumer relief is just too uncertain to be accrued now. Incidentally, if this agreement in principle happened in January, why does it reduce income for the quarter ended last December? Conveniently, I have written before about this very issue. The rule is roughly that if you have a loss that accrued in the past (say, because you did bad mortgage stuff in 2005-2007 and everyone knew about it by certainly 2015 at the latest), and if you become able to reasonably estimate the amount of the loss after a quarter ends but before you issue the financial statements for that quarter, then you can include the loss on the financial statements for that past quarter.
Goldman hasn't put out its fourth-quarter 2015 financials yet, so it gets to include this loss in those financials. None of this matters all that much, incidentally. There are people who believe that everything bad in financial markets is due to naked short selling, and that the big banks are in a massive conspiracy to facilitate illegal short selling to manipulate the stock prices of disfavored companies. But actually existing naked-short-selling conspiracies are usually about saving some money on stock-borrow costs, not about price manipulation, and there is even an argument that allowing naked short selling could be good for markets. Never mind that now, though. The relevant point here is just that, if a customer wants to do a short sale
, her broker needs to "have reasonable grounds to believe that the security can be borrowed." That's based on the fact that "Over the course of the relevant period," i.e. November 2008 to mid-2013, "the number of locate requests that pended to the Demand Team grew significantly, reaching more than 20,000 locate requests per day at its peak." That's 20,000 requests per day beyond Goldman's automated system, which handled "the vast majority" of locate requests. It's so efficient! The “fill from autolocate” function enabled Demand Team members to select as many of the locate requests routed to it from Goldman’s automated model as desired (for certain customers, 3,000 to 4,000 locate requests were sometimes selected at one time) and hit the “F3” key to process the selected requests with one keystroke. For example between 6:45 a.m. and 7:45 a.m. each day, on average, the Demand Team filled locate requests for several thousand unique CUSIPs. Additionally, in one instance, a Demand Team member granted over 5,486 locates in a 35-minute period.