Knight Capital Group remains under scrutiny more than three months after going live with improperly installed software that triggered a wave of errant trades that nearly put the firm out of business.
The Aug. 1 incident led to the $389.9 million third-quarter loss that Knight Capital posted in October, drove the firm to hire IBM to review its software and pushed the Securities and Exchange Commission to convene a roundtable of industry experts to discuss ways of preventing such market dislocating errors.
In the wake of that episode, the buy side began taking a closer look at the algorithmic trading tools it gets from third-party providers like Knight Capital. There's greater urgency among buy-side firms to figure out how their algorithms could be impacted by the sort of errant orders that originated that fateful day from Knight's market-making unit, which executes order flow for retail investors.
But as the firm looks to move on, Brendan McCarthy, head of sales and trading for Knight's algorithmic trading group, wants to set the record straight about his side of the business. McCarthy took over the role in September following the departure of Joseph Wald, the first senior executive to leave the company after the software glitch hit.
"It had nothing to do with any of our algorithmic trading," McCarthy says of what occurred on Aug. 1. Knight Capital's algorithmic trading group is "completely separate" from the market-making unit, he says. "We don't share any of the same technology. We don't even share the same resources. We have a completely independent team of 40 people in development, and none of them are shared or do any work between the two businesses."
Knight Capital's businesses include its market-making unit, where the software glitch emanated from; its electronic execution services unit; and its institutional sales and trading operation. McCarthy, whose group is part of the electronic execution services unit, says the Aug. 1 glitch hasn't damaged relationships with clients, which are mostly buy-side firms and some sell-side ones.
"We really haven't had any clients who've told us, 'You know, we're not coming back,'" McCarthy says. "We had clients who wanted to take their time, clients who wanted to do their due diligence, wanted to talk to some people, get a lot of information out of us to get comfortable with things. But generally our clients have come back and been very receptive and supportive of us."
Step By Step
One question that comes up from clients is how Knight Capital's trading algorithms are tested in the aftermath of Aug. 1. McCarthy contends that before a strategy is allowed to go live, it must first endure a "rigorous" development phase in which there's no interaction with the marketplace.
After initial testing and analysis, once his team is comfortable that an algorithm is ready for "production" testing, it then does a very limited test run from the trading desk, McCarthy says. With this round of testing, the group's client service and trade support teams must vet the strategy.
At this embryonic stage, McCarthy says, his team tries out algorithms on a limited basis. This involves the nascent algo being used to make one order, maybe for 100 shares. After that, "we would do a thorough analysis around the expected marketplace interaction and the actual marketplace interaction," he says. "All the while that 100 shares is executing, we would have the trade support team watching the order placements, the outbound message traffic and the inbound message traffic to make sure nothing went awry."