One lesson from the financial crisis is that innovation on Wall Street almost always seems to outpace regulation. The explosion of derivatives such as credit-default swaps, the peddling of dubious mortgages, and abuse of off-balance-sheet vehicles by banks to mask risk all caught regulators napping.
It could happen again — and soon. Just as regulators get their arms around high-frequency trading, in which ultrafast computers dart in and out of securities at lightning speeds, dangerous new strategies are springing up. A recent New York conference for investors and technology providers showcased the latest techniques being devised by quants — Wall Street mathematicians valued for their quantitative skills. The chatter at the conference was about hot new tools ranging from artificial intelligence and ultrafast computers to mine data to the use of social networks — Facebook, LinkedIn and, yes, Twitter.
The show stealer: artificial intelligence — specifically a brand known as machine learning, in which powerful programs crunch reams of data to identify patterns and predict future outcomes — is on the cusp of becoming a major driver of profits on the Street.
Spencer Greenberg, the conference keynote speaker whose small New York hedge fund, Rebellion Research, uses artificial intelligence to choose stocks, said artificial intelligence “won’t get a computer to learn to speak to the CEO, but it can get it to uncover fundamental principles of investing.” By searching through reams of historical data, these sophisticated computer programs can spot investment trends and on their own come up with strategies. Computers can “learn” that stocks with low price-to-earnings ratios compared with earnings growth often rise more than other stocks, for example.
In the past, AI largely disappointed when tested in the real world. Now, Google deploys AI techniques to hunt out websites that correlate with search terms. Netflix and Amazon use AI to recommend movie matches for users based on previous selections and preferences. And the IBM supercomputer Watson defeated two elite players on Jeopardy.
As more Wall Street firms use machine learning, however, there are risks. Many are using the technique to predict very short-term moves in the market — often a minute or less. What happens if many firms using similar techniques predict a sharp downfall and suddenly sell their holdings — making the decline even worse?
The impact can be quick and dramatic. On Sept. 8, 2008, for instance, a six-year-old news story about a United Airlines bankruptcy was posted on the Internet, triggering instant selling of the stock by computers programmed to read news feeds. The company lost about three-quarters of its market value as the stock plunged from just over $12 to $3 a share, before someone realized it was old news. The company had emerged from bankruptcy is 2006.
Michael Kearns, a computer professor at the University of Pennsylvania who specializes in the use of AI as an investing tool, says most firms that employ machine learning use the same data to build their models. “One danger is that we all arrive at the same strategy, we’re all looking at the same thing,” he says. “The data correlates your behavior.”
Meanwhile, Johan Bollen, a professor at Indiana University’s School of Informatics and Computing, has designed a computer system to track and analyze posts from Twitter. Using computers to hunt for sentiment indicators, such as fear or confidence, Bollen says his models can predict the direction of the market several days in advance with an astonishing 87 percent accuracy. When his system detects a spike in worry on Twitter posts, for example, he found that the Dow Jones Industrial Average is likely to drop in the near future. A burst of confidence indicates an upward move. While the use of such strategies for now is likely harmless, the risk is that many firms will start to use them, wagering billions based on the same data.
The risk of a sudden correlated short-term move driven by computers was apparent in the May 6, 2010 “flash crash.” The Dow lost nearly a thousand points in a matter of minutes, in part due to erratic behavior by computer algorithms, before recovering that same day.
As computer trading spreads around the globe, some say the chances of unpredictable events such as the flash crash are growing. John Bates, chief technology officer at Progress Software, a Bedford, Mass., trading-software firm, has warned of a “splash crash” in which chaotic trading spreads between currencies, stocks and bonds.
Trading outfits such as Rebellion pose little immediate threat to the market, because their investments are long term and they aren't likely to make sudden destabilizing moves. But as more firms use machine learning for long-term investing and their strategies become highly correlated, it could lead to broad, self-reinforcing swings in the market, either up and down.
The Securities and Exchange Commission and Commodities Futures Trading Commission have spent the past 10 months investigating the flash crash. That has given them reams of new information about how the market works, including insights into the practices of many high-frequency trading firms, officials at the regulatory agencies say. With that head start, regulators could press firms for more details about how they build their models. The SEC declined to comment for this article.
The SEC has proposed a rule to force organizations such as the Financial Industry Regulatory Authority, the self-policing body that oversees U.S. securities firms, to build a so-called consolidated audit trading system that would track orders across securities markets. Such a system would “allow us to rapidly reconstruct trading activity and quickly analyze both suspicious trading behavior and unusual market events,” SEC Chairwoman Mary Schapiro said in written testimony for a recent Senate subcommittee hearing. Estimates of the cost of such a system have varied.
But Paul Zubulake, a senior analyst at Aite Group, which tracks trading technology, says regulators need to hire expert help from within the financial industry to create systems to automatically hunt for trends — or nefarious behavior. “Regulators should look into the black box and see what the strategies are,” he says. “But they shouldn’t get caught up in creating their own [system]. They need to outsource it to those firms that have the technology.” (His firm would not be among them.)
The Future of Finance is Getting Closer (iStockAnalyst)
Trading Robots, Automated Trading and Futures Trading (Counting Pips)
Earthquakes and Flash Crashes – The Coming Mini Flash Crash (Business Insider)