Computers read news, and trade on it quickly

NEW YORK (Reuters) - It takes a person about 10 minutes to read a 2,500-word, front-page feature story in the Wall Street Journal. Computer programs increasingly being used by investors to parse news stories can process one in about three one-hundredths of a second.

Traders work at a kiosk selling shares of the Blackstone Group LP at the New York Stock Exchange in New York, June 22, 2007. It takes a person about 10 minutes to read a 2,500-word, front-page feature story in the Wall Street Journal. Computer programs increasingly being used by investors to parse news stories can process one in about three one-hundredths of a second. REUTERS/Keith Bedford

Algorithms -- problem-solving programs based on mathematical formulas -- are making it easier for investors to filter the massive amount of text produced by news wires, newspapers, industry journals, clinical studies, and legal filings for kernels of information, and trade on them in the blink of an eye.

Though the expanding array of news on nontraditional media like blogs and chat pages is a challenge for the robot readers, the speed and efficiency offered by news mining algorithms are helping hedge funds with just a handful of staff generate as many trades as a giant investment bank and becoming a potential boon to the media industry.

“This is a new class of information technology,” said John Partridge, vice president of industry solutions with StreamBase Systems, a technology provider that specializes in processing and analyzing real-time streaming data.

High-frequency investors such as hedge funds are using news mining platforms like those offered by StreamBase to troll through thousands of electronic feeds of streaming text to identify key phrases on which to trade.

Popular phrases include “lowers its outlook” or “raises guidance” or even buzzwords like “stellar performance” that could potentially push a stock lower or higher.

Hedge funds, with their rapid-fire trading style, often allow the news mining platforms to make trades on their own, capitalizing on the technology’s speed.

However, longer-term investors are less interested in flooding the market with orders after a particular headline. They are using the platforms to keep track of developments that may affect companies in their portfolios or influence their strategies, technology developers said.


News mining is not just for stock trading, either. For example, French investment bank BNP Paribas’ “weakness indicator” counts the number of times the words weak, weakness or weakening are used in the Federal Reserve’s Beige Book report on regional U.S. economies.

More than 50 references in a report typically signals the economy is on the brink of a recession.

Hedge fund investors familiar with news mining technology said an algorithm based on the “weakness indicator” could easily be created to sell dollars and U.S. stocks and buy bonds if more than 50 references were found.

“What the machine is looking for is the same thing that the human is looking for. It can just find it more quickly,” said Richard Brown, business manager of NewsScope, a company owned by Reuters Group Plc that produces machine-readable news.

Rather than just highlight words or phrases, some of the most sophisticated news mining platforms can take multiple strands of news from wire agencies and Web sites and score the significance of various items.

For example, headlines from a reputable news organization with the words “Middle East,” “tension” and “hostility” would be given a higher score, especially if oil prices are rising, than an anonymous blog entry with the same key words.

The same headlines would be given an even higher score if other reputable news agencies carried similar stories.

“A lot of times, the content that’s important is not in a single article or document,” said David Leinweber, a financial technology consultant with Leinweber & Co. “The idea of considering individual news stories only as atomic events misses some things,” he said.

On his own blog “Nerds on Wall Street,” Leinweber noted the example of Accentia, a pharmaceutical company whose share price shot up 70 percent one morning in October 2006 after the successful trial of a human cancer vaccine was announced in a press release.

However, the press release was based on an article from a medical journal published a month earlier. Also, local press in St. Louis, where Accentia has a plant, reported on the testing a week before the press release, and a blog for patients discussed the drug days before the stock jump.

An investor using news mining technology could have been buying into the company days, if not weeks, before the big share price rise.


Computers, however, are not perfect when it comes to reading the various forms of language in both standard and nonstandard media.

Consultant Leinweber added that machines often have difficulty with subtle double negatives and vague pronouns that human readers can understand easily with context.

For example, machines could potentially stumble when it comes to a sentence such as: “The company’s chief executive said he did not dislike the way that that product sold well there.” A person could scan the sentence and understand it.

The growing amount of text and information available on blogs, chat rooms and online forums also pose challenges to robot readers.

“That’s one of the limitations. When you look at chat room and blog content, it’s the emoticons, it’s the profanity, it’s sarcasm or all caps,” said NewsScope’s Brown.

Still there is growing interest in the investment community in being able harness the information available in so-called social media.

Darren Kelly, senior vice president at Collective Intellect, a company that specializes in filtering and ranking media content, said blogs and online forums can provide a unique window on sentiment surrounding an issue or a stock.

“The usual multiscreen setup that everyone has used in finance for the last 20 years no longer gives them all the information that’s available,” Kelly said.