Financial Trading by Machines that Capture "Emotional Context"

Here is an interesting article about the new frontier of automated trading. Algorithms tracking and reacting to market moves are no longer ‘fast enough.’ Increasingly, algorithms aggregate raw sentiments from newspapers and blogs and issues orders based on them, bypassing human interpretation.

The system may indeed be at the cutting edge, but it is dangerously susceptible to easy manipulation. You can imagine the next wave of robo-crawlers artificially pumping up a news story designed to highlight a false vulnerability that will depress shares of a sector or stock that the manipulator shorted. Very dangerous.


September 14, 2008

Ideas & Trends

I Got the News Instantaneously, Oh Boy


It’s a tantalizing daydream for horse-track bettors, bookmakers and Wall Street traders: the thud of tomorrow’s newspaper landing on the doorstep today.

While no one has figured out how to make that happen, Wall Street’s computer scientists and linguists keep trying to find quicker ways to react to the news by creating ever-more complicated algorithms, the mathematical formulas that execute stock trades automatically based on such criteria as headlines and news stories. The idea is to buy or sell at a faster clip than the guy or computer at a rival trading desk.

All this can go horribly wrong, as United Airlines learned last week when a six-year-old story about the company’s 2002 bankruptcy filing gained new life on the Internet, triggering a cascade of stock sales.

In a matter of about 12 minutes more than $1 billion in stock-market value evaporated.

Human error seems to have played only a minor role. The financial damage was mostly the result of the interplay between the algorithms that search and compile information from the Web and the ones that Wall Street firms and hedge funds use to make trades automatically.

“I think it underscores the financial arms race that trading has become,” said Prof. Andrew W. Lo of the M.I.T. Sloan School of Management. “The downside of the technology is occasional glitches like these.”

Witness another recent case that had the potential to cause a stock market wipeout, but benefited from serendipitous timing: after the close of trading on Aug. 27, Bloomberg News inadvertently released an obituary of Steve Jobs, the chief executive of Apple — who, despite frequent rumors of ill health, was, and is, very much alive. The story was quickly retracted.

In the case of United, a spokesperson for the airline said it was still investigating what happened. But experts in the growing field of news-driven trading systems point to the case as a cautionary tale. Once the price of United’s stock started moving downward, other robotraders spotted the trend, and acted. “Every other algorithm was then trading on price movements,” said Richard Brown, the global business manager for machine readable news (yes, that’s his official title) at Thomson Reuters.

Automatic stock trading based on algorithms that react to market moves and conditions have been a feature on Wall Street for years — nearly 50 percent of all trades on the New York Stock Exchange are expected to be automated by 2010, according to the Aite Group, a Boston-based consulting firm, up from roughly 30 percent in 2006. Robotrading based on news events is of more recent vintage.

“It’s growing exponentially, but nobody wants to talk about it because they feel they have an edge,” said Professor Lo, who directs the M.I.T. Laboratory for Financial Engineering. The formulas are viewed by banks and hedge funds as trade secrets, he said.

Even the news organizations themselves are in on it, developing support tools so traders can more quickly mine their news content to make trades. Professor Lo has been collaborating with Thomson Reuters to develop news-based algorithms. Dow Jones and Bloomberg offer similar services.

Many of the news-based trading formulas today have as their genesis research Professor Lo conducted at M.I.T about five years ago when he sought to measure the “emotional context” of stories in The Wall Street Journal.

Professor Lo chose words such as “anxiety” and “depression” and counted how many times they showed up in the newspaper on a particular day — and then looked for correlations to the stock market.

“How many times they appeared on a given today tended to be a leading indicator of the S. &P.,” Professor Lo said. From there, he developed more complex methods such as the ordering of words — pairs and triplicates. “We were among the first to use algorithms to read the news,” he said.

The formulas assign a weighted value to certain words — “bankrupt” may be given a higher score than, say, “anxiety” or “down” — and when a certain level is reached, depending on how the model is programmed, it will automatically initiate a trade.

The important factor is speed — receiving the data, analyzing the information and making a trading decision in a matter of milliseconds.

It doesn’t stop with hard news. Services exist that track sentiment for a company on blogs and social networks on the Web. “We help companies listen to and monitor social media,” said Tim Lefkowicz, the president of Collective Intellect, which works with hedge funds and corporations.

Eliminating the human touch from the process seems to be what wiped out all that value in United’s stock — because any person who follows the company or owns the stock likely would have known to dismiss the bankruptcy report as old news.

“If you put a person in, it catches a lot of errors, but it slows you down,” said David J. Leinweber, the founding director of the Center for Innovative Financial Technology at the Hass School of Business at the University of California, Berkeley.

Here is the rough version of events. At 1:36 a.m. E.D.T. last Sunday, Sept. 7, Google’s search “crawler” picked up a 2002 news article about United filing for bankruptcy from the Web site of The South Florida Sun-Sentinel; for some reason the outdated story had been listed on The Sun-Sentinel’s list of most popular business stories. (United emerged from bankruptcy protection in 2006.)

The next morning, an employee of the investment advisory firm Income Securities Advisors saw the story and posted it to the company’s own wire service, which is available over Bloomberg’s trading terminals. United’s stock plummeted soon after.

In a statement, the Tribune Company, which owns The Sun-Sentinel, said the bankruptcy story “contains information that would clearly lead a reader to the conclusion that it was related to events in 2002. … It appears that no one who passed this story along actually bothered to read the story itself.”

Mark Palmer, president of Streambase Systems, which designs software for automated trading, said: “What happens is the first crack in the ice is the news gets fed out to the algorithm, and that triggers it to sell. Someone puts in a big order and that gives the signal to other algorithms. Then another system picks up the sell order. That’s the whiplash. The problem is there’s a cascade effect.”

Digg This
Reddit This
Stumble Now!
Buzz This
Vote on DZone
Share on Facebook
Bookmark this on Delicious
Kick It on
Shout it
Share on LinkedIn
Bookmark this on Technorati
Post on Twitter
Google Buzz (aka. Google Reader)

related posts

post a new comment