Technology, Data and AI

On 1 August 2012 the New York Stock Exchange opened a new venue for retail orders, the Retail Liquidity Program, which the SEC had approved a month earlier as a twelve-month pilot. Knight Capital, the Jersey City wholesaler whose trading was then about a tenth of all trading in listed US equities (SEC order), wanted its broker customers' orders to be eligible from the first day. Its engineers wrote new code for SMARS, the firm's high-speed router that split large "parent" orders into child orders and sent them to outside venues, and deployed it in stages from 27 July. The new code replaced a function called Power Peg, discontinued in 2003 but still present and callable, and it reused the flag that had once switched Power Peg on. One technician did not copy the new code to one of the router's eight servers. No second technician reviewed the deployment, and there were no written procedures requiring one (SEC order, 16 October 2013).

From 8:01 am, before the open, an internal system sent 97 automated emails referring to SMARS and "Power Peg disabled" to a group of staff. They were not designed as alerts and nobody acted on them. At 9:30 the orders flagged for the new program reached all eight servers. Seven handled them correctly. On the eighth, the flag woke Power Peg, which no longer had a working check that it had filled its parent order, because in 2005 the function that counted fills had been moved and Power Peg never retested. The server sent child orders continuously. Trying to fill 212 customer orders, Knight obtained more than 4 million executions in 154 stocks, for more than 397 million shares, and built about $3.5 billion of unwanted long positions and about $3.15 billion of unwanted shorts (SEC order; the SEC's word is executions, not orders). Searching for the fault in live trading, staff uninstalled the new code from the seven correct servers, which spread the fault to all eight. There was no firm-wide capital threshold linked to automated controls that would have stopped order entry, and no procedure for disconnecting a malfunctioning system from the market.

It took about 45 minutes. The SEC put the loss at more than $460 million; Knight's own early estimate was about $440 million pre-tax. The firm had earned net income of about $115 million in 2011 (company figures, reported). A $400 million rescue led by Jefferies kept it trading at the cost of most of its equity (Chapter 7); it merged with Getco to form KCG Holdings on 1 July 2013, and Virtu Financial bought KCG on 20 July 2017. On 16 October 2013 the SEC fined Knight $12 million, its first enforcement case under the market access rule, Rule 15c3-5, which requires broker-dealers with market access to have pre-trade risk controls (SEC release 2013-222). Chapter 7 tells Knight's story as a market-making business. This chapter tells it as a technology failure, because nothing about the market caused it.