Goldman puts mortgages and naked shorts behind it: Matt Levine

by MPA19 Jan 2016

The basic rule is that you're not supposed to do "naked short selling." Short selling is selling stock that you don't own: You borrow the stock, sell it on the exchange, deliver the borrowed stock, and hope that it goes down before you have to buy it back. Naked short selling is selling stock that you haven't even borrowed: You sell stock on the exchange, don't deliver it, and when your broker demands that you deliver it you hang up on him. That's illegal.

Or sort of illegal. The way the rule actually works is that if you want to sell short, you have to tell your broker that you are selling short, and then your broker has to "have reasonable grounds to believe that the security can be borrowed." (This checking with your broker is called a "locate.") You don't have to actually borrow it to sell it short, and neither does your broker. Your broker just has to think he can. If it turns out after the fact that your broker actually can't borrow the stock, you have a "fail," and your broker has to quickly find stock to borrow or, if he can't, close out your short position.

Goldman has a lot of customers who want to do short sales. In fact, it gets at least tens of thousands of requests for locates a day. So it automated the process:
Goldman employed a system where the vast majority of customer short sale locate requests were handled by an automated model that would either grant, in whole or in part (or “fill”), deny, or route (or “pend”) the requests for further review to the Demand Team, a group of ten to twelve individuals who worked on Goldman’s securities lending desk. The automated model would review and fill locate requests based on certain available inventory reported to Goldman by certain lending banks and brokerages that fed into Goldman’s automated model at the start of each day after being reduced by Goldman based on their experience with various lenders (the “start-of-day inventory”). As the automated model processed locate requests, it reduced that start-of-day inventory on a 1:1 basis for all shares that were used to grant locate requests (regardless of whether the client actually used the locate). When the start-of-day inventory was depleted in that manner, the automated model would pend subsequent locate requests to the Demand Team for further review and processing.

Fine. The computer knows how many shares it has, and as it gets requests, it crosses out those shares until it has none left, at which point it sends requests to the humans. But eventually the humans were getting "more than 20,000 locate requests per day," which is too many for humans. So:

read more > 1 2 3 4



Is TILA-RESPA a good or bad thing long term?