Investment and trading firms must impose more discipline developing quantitative models to avoid mishaps that can harm investors, according to a Securities and Exchange Commission official.
Reliance on mathematical models to make trading decisions requires firms to ensure that automated systems operate as planned and are properly tested, Erozan Kurtas, an assistant director in the quantitative analytics unit within the SEC’s Office of Compliance Inspections and Examinations, said at a conference about high-frequency trading yesterday.
The SEC is emphasizing the responsibilities of firms using automated strategies to invest or trade in the wake of events such as Knight Capital Group Inc. (KCG)’s technology mistake that cost the firm more than $450 million in August, and a coding flaw that led to $217 million in client losses at a unit of Axa SA a few years earlier. The Axa case was the first targeting an error in a quantitative investment model, according to the commission.
“If we don’t sort this out, we’re going to have lots of Axas or Knights,” Kurtas said.
The OCIE unit, formed earlier this year, uses quantitative analytics to understand how companies manage risks posed by technology and mathematical models, Kurtas said. OCIE administers the SEC’s national examination program of companies such as investment advisers, brokers and exchanges.
Some firms don’t do enough to test and manage their software, Kurtas said. Individuals at some investment companies don’t know the difference between the simulation of potential events and back-testing to see how a model would have worked under past market conditions, he said. There’s also insufficient testing for incorrect data.
“There are so many companies out there who are not properly testing for stale data,” which can affect how models function, he said. “Sometimes people tested for certain periods and ignored certain other periods,” Kurtas said. “You can’t do that,” particularly if the firm’s marketing materials say the strategy works in all market conditions, he added. “You have to make sure the model works the way you are representing, otherwise it’s fraud.”
Testing should also be conducted by individuals who didn’t write the computer code, said Kurtas, who has a masters of business administration degree from Carnegie Mellon University and a doctorate in electrical and computer engineering from Northeastern University. He previously was a senior analyst at Standard & Poor’s who evaluated collateralized debt obligations and funds of hedge funds and earlier was a research director at Seagate Technology Plc, where he led a group designing algorithms to identify anomalies in data sequences.
Knight dodged bankruptcy in August after losing more than $450 million in a trading error that resulted from the installation of new software. The glitch caused the Jersey City, New Jersey-based market maker to unintentionally spew orders into the market. Knight was bailed out by six financial firms a week later and is now considering takeover bids by two rivals.
Axa Rosenberg Group LLC, a unit of the French insurance company, lost $217 million because of a malfunction that was kept from the unit’s CEO and investors, the SEC said in February 2011. The error, introduced in 2007, was discovered in June 2009. The firm was censured and agreed to a $25 million penalty in addition to a payment of $217 million to cover client losses.
The units “failed to disclose the error and its impact on client performance, attributed the model’s underperformance to market volatility rather than the error, and misrepresented the model’s ability to control risks,” the SEC said. Barr M. Rosenberg, the division’s co-founder, separately agreed to pay $2.5 million to settle claims of securities fraud.
Companies need “proper documentation of all models” and some compliance personnel should be able to read the source code to understand how the systems work, Kurtas said. Changes to models should be tracked, there should be a process to escalate emerging problems to executives, and the risks of quantitative systems should be disclosed to investors, he said.
Firms that don’t document how their models work will get longer visits by SEC officials, Kurtas said. If the source code for the model is the only documentation, “we have to look at the source code,” he said.
The SEC’s OCIE program had net costs of $236 million in the year ending Sept. 30, the highest outlay of funds after the enforcement division’s $401 million, according to a report by the Government Accountability Office last month.
The Managed Funds Association, which represents hedge funds and managed futures firms, urged the SEC to consider requiring every broker to form a technology advisory group “to review or perform its pre-technology release reviews or to develop technology risk assessments,” according to an August letter that followed Knight’s blunder.
The commission could require every firm to have a technical manager certifying to senior executives that the broker conducted appropriate testing and market simulations of software releases and complied with its policies for installing programs, Stuart Kaswell, general counsel at the Washington-based MFA, wrote. The group urged the SEC to consider updating the market- access rule implemented last year mandating that brokers to conduct credit and risk checks on orders before they’re submitted to exchanges or other venues.
It said the obligations should be imposed on individual rather than aggregated orders to flush out technical glitches.
The traditional separation within brokers of the revenue- generating equities-trading business and technology-support division is “outdated and dangerous” in a faster, automated market and these groups should become more integrated, David Weisberger, a former equities executive at Citigroup Inc., told the SEC in an October letter.
Since technology errors can cause big losses within a firm and harm investors in the broader market, risk controls should be incorporated into business decisions, said Weisberger, who previously was managing director and head of global electronic market access at Citigroup and is now executive principal at market-maker Two Sigma Securities, an affiliate of Two Sigma Investments LLC, which manages about $10 billion.
Automated monitoring systems of algorithms or computerized trading strategies should be overseen by human beings and operate independently of a firm’s trading software, Weisberger wrote. Risk managers and technology experts should get real-time alerts that flag potentially aberrant activity so they can take immediate action, he said.
“Whether a risk threshold is breached by software malfunction or mistakes made by a human trader, the system should alert these individuals so that predetermined escalation procedures may be followed,” Two Sigma told the SEC in the letter. “The lack of such a clear command-and-control process could allow errors to compound and, ultimately, threaten the integrity of the market as a whole.”
Technology’s role in trading should drive brokers to improve the way they write computer code and deploy software, including how strategies are simulated before they’re used and what happens when activity looks abnormal, Weisberger said. What once was seen as the traditional back office, which includes technology, operations and accounting, should occupy a more central role in business decision-making, he said.
Compliance employees at investment advisers and brokers may need to acquire new skills in financial engineering and computer science, Kurtas said. If some compliance staff can’t read source code, “we don’t believe they can put together a robust compliance system,” he said. Programmers often design systems without sufficient knowledge of regulations, he added.
The SEC is building up its market-analysis and oversight staff by adding trading specialists and individuals with doctorates in mathematics, financial engineering, statistics and computer science, he said. The commission is also implementing a system that collects order and transaction data from the real- time feeds exchanges sell to high-frequency traders and brokers to gain access to tools it’s never had, he said.
“We don’t have the budgets that the private sector has,” Kurtas said. “We’re not going to be able to hire armies.” Still, smarter inspections and oversight of regulated companies is possible. “We are just looking for a few good Jedis,” he added, referring to a group of guardians that fought for justice in the 1977 “Star Wars” movie.
To contact the reporter on this story: Nina Mehta in New York at email@example.com
To contact the editor responsible for this story: Lynn Thomasson in New York at firstname.lastname@example.org.