NYSE Fines Credit Suisse For Algo Run Amok
January 13, 2010
Amidst concerns of trading algorithms running amok, Credit Suisse has been fined $150,000 by New York Stock Exchange (NYSE) Regulation for poorly supervising the development and execution of an algorithm that clogged the exchange’s order processing system and severely delayed messaging traffic on five NYSE-floor trading posts.
The NYSE action was announced several hours before a Securities and Exchange Commission hearing would be held Wednesday to decide whether to pursue proposals requiring additional risk management controls and re-examining equity market structure in light of high-frequency trading’s rapid growth.
In determining the fine--petty change for a major Wall Street firm—NYSE cited mitigating factors such as the significant enhancements to supervisory policies Credit Suisse has made since the exchange uncovered the incident in January 2008 and its cooperation in NYSE’s investigation. Nevertheless, regulators tend to fire initial warning shots and levy much more significant penalties on subsequent infractions.
NYSE’s action illustrates the concerns that have emerged in recent years about traders losing control of algorithms and flooding the exchanges with unintended orders.
In this case, the trade did not affect any Credit Suisse customers. The algorithm in question, known as SmartWB, was being used in the company's proprietary trading operation.
The problem emerged when one of the firm’s proprietary traders manually changed the limit prices for unexecuted orders on his own accord, resulting in an electronic loop that generated hundreds of thousands of unintended messages sent to NYSE.
“It highlights the risks in electronic trading where development is many times done directly in a production environment - the traders are developers, and thus normal development protocols and thorough testing may not be effectively done,’’ said John Jacobs, director of operations at Lime Brokerage.
Credit Suisse has become a leader in algorithmic trading since the launch of its Advanced Execution Services unit in 2001.
Credit Suisse declined to comment.
According to the NYSE account, 20 minutes before the market close on November 14, 2007, SmartWB began trading on 129 out of 800 securities it examined and generated execution orders.
After 45 seconds, the trader, who also function as a programmer, changed the trading parameters to set new limit prices on existing orders. However, when he double-clicked an arrow to generate cancel and replace messages for all unexecuted orders with new limit prices, the brief period between clicks resulted in generating a separate set of "cancel" and "replace" requests before replacing the unfilled orders.
That created a loop resulting in SmartWB sending 600,000 messages related to seven securities to NYSE during the 20-minute period. The NYSE responded with 405,000 “reject-unmatched cancel” messages. That volume caused messaging traffic in five trading posts to slow or stop completely, “…and trading in each security at these posts could not be completed by their normal closing time of 4 p.m.” These "didn’t close until as late as 4:27 pm.,” according to the NYSE-action report.
NYSE notes that Credit Suisse never implemented a circuit breaker or similar safeguard to ensure such events didn’t occur and the SmartWB algorithm was not designed to alert users to rejected messages or unusually high message traffic.
“Specifically, the Firm violated NYSE Rule 401 by implementing a proprietary algorithm that did not have appropriate checks designed to prevent the continuous submission of erroneous messages or to alert the Firm in the event of rejected messages,” NYSE states.
NYSE also notes that Credit Suisse violated Rule 342 for failing to supervise properly the development, testing and rollout of proprietary algorithms such as SmartWB. In fact, the trader reprogrammed the algorithm without seeking approval to allow manual changes to limit prices.
Tabb Group analyst Adam Sussman notes the “hype” around the potential for algorithms to wreak havoc, and that Credit Suisse’s problem was actually prompted by human interference.
“It is quite revealing that … the root cause of the November 2007 incident was the introduction of manual functionality that allowed the user to override certain parameters,” Sussman says.
The problem stemmed from a simple “fat-finger error,’’ Brian Hyndman, senior vice president of NASDAQ Transaction Services, said Wednesday morning at the Capital Markets Consortium breakfast in New York. He said this was based on discussions he had held with the firm in question.