Louis Morgan is a news junkie. As managing director of HG Trading, a small hedge fund based in Hartland, Wisconsin, he continually scans online news wires in search of the latest tidbit that might move a stock. But lately he’s finding it harder to keep up with the competition.
Over the past year wire services have begun targeting a select group of clients -- bank trading desks, hedge funds and other trading groups -- offering computer-ready news feeds that can be routed directly into algorithms. Machine-readable news is coded for a computer, making it faster to process and run through algorithms than text is. Investment companies that use the service can get a few fractions of a second advantage over the rest of the market.
Morgan, a blustery risk-taker who lists “global domination” as his career objective on his résumé, figures he can’t afford not to buy the service. Keeping up with trading technology has been a lifelong pursuit. After receiving a bachelor’s degree in engineering from the University of Notre Dame in 1958, he quickly moved into finance with a series of jobs in money management and trading, including a stint as an options trader in Chicago, where he was a charter member of the Chicago Board Options Exchange from 1973 to 1986 and worked on the first online options analysis system. In 1980 he founded and served as chairman of Chicago-based PC Quote, an early online stock quote provider.
“If you have a news item that is exciting to you regarding a specific stock, and your computer determines that and enables you to be the first to grab it, you win,” says Morgan, who joined HG Trading in 2002. “If you are second, you might be buying from the guy who just bought it and end up paying up. By the time you read the wire, it is too late.” HG runs a statistical arbitrage hedge fund with less than $10 million under management, and Morgan, its primary trader, is working to adapt the firm’s algorithms for computer-ready news feeds by year-end. He’s not the only one embracing the technology.
Barely a year old and expanding rapidly, machine-readable news is offered by such news and data providers as New Yorkbased Dow Jones & Co., Stamford, Connecticut’s Thomson Corp. and London-based Reuters Group. Data providers have been adding and developing computer-ready feeds providing information on economic index reports, corporate earnings and mergers and acquisitions. News is tagged with computer code indicating key topics from each story, which can then be used in an algorithm. Providers have also started to tag archived news and send it out in computer-ready formats.
“We are seeing enormous demand for these types of products from most major clients, as well as the midtier ones,” says Andrew Meagher, manager and director of news and strategy at Thomson Financial News in New York. “Clients seek an edge that allows them to differentiate their trading strategies.”
Andrew Lo, a finance professor at the Massachusetts Institute of Technology’s Sloan School of Management and director of the MIT Laboratory for Financial Engineering in Cambridge, Massachusetts, is helping advance the technology and trying to capitalize on it. Through his work at the university, Lo has been advising Reuters on the development of that company’s tagged news products. He is also the founder and chief scientific officer of AlphaSimplex Group, a Cambridge firm that manages $550 million in hedge funds and has begun incorporating news into its algorithms.
“The pace at which this technology is developing is striking,” Lo says. Though the hedge fund firms relying on news-sensitive algorithms are tight-lipped about their usage, Lo estimates that roughly 100 hedge funds, and 25 or so other portfolio managers, have already adopted the technology.
Until now, writing algorithms that recognize news-based market patterns was extremely labor intensive. It involved collecting years of archived stories and turning them into data, then analyzing the data for patterns of market activity, which then needed to be incorporated into investment decisions. Only a few of the biggest hedge funds and trading desks had the necessary resources and, even then, the process was usually too slow to provide a trading edge. Nonetheless, some of the larger firms have been experimenting with news-driven algorithms for the past few years.
“Before, a firm would have had to have legions of people looking at news, trying to provide their own read on it,” says Mark Palmer, a vice president and general manager of Bedford, Massachusettsbased Progress Software Corp.'s Progress Apama algorithmic trading platform, which analyzes and detects patterns within the news in milliseconds. “By the time they did that, it was too late to take advantage of a trading opportunity.”
The potential demand for such services has not been missed by the major news providers, which have been rushing to beat their competitors to market with offerings.
Dow Jones started experimenting with computer-ready feeds of live economic reports in July 2006 -- initially working with U.S. Labor Department monthly indicators. In March the company launched its computer-ready service, Elementized News Feed, which provides feeds on 150 economic indicators. And this month it announced a deal to offer Dow Jones’s computer-ready news feed on Progress Software’s Apama algorithmic trading platform. Dow Jones, which has applied for a patent on its technology, is also testing computer-ready feeds of corporate earnings and M&A news.
Reuters began selling computer-ready news feeds through its NewsScope service in December 2006. In April it beefed up its offering by purchasing ClearForest, a provider of computerized text solutions that include a program for automated tagging of text.
In May, Thomson Corp., which began selling its computer-ready feeds this year, agreed to acquire Reuters in a deal valued at $17.2 billion. The merger, which has yet to clear regulatory hurdles, would create the world’s biggest source of financial data, instantly making Thomson a major provider of computer-ready news.
The pricing and complexity of computer-ready news feeds varies, as software can be tailored to individual clients’ needs. “It can cost anywhere from a few thousand dollars a year to several million dollars per year, depending on what you want to do,” says Richard Brown, business manager for Reuters NewsScope in London. Adds Robert Prinsky, executive director and senior editor of Dow Jones’s institutional product development: “We’re priced where we don’t need many clients.”
Given the historical use of news by trading firms, it’s not surprising that they have closely monitored research done by MIT’s Lo and other academics specializing in linguistics, artificial intelligence and market dynamics.
News services have grown interested in providing computer-ready news feeds because of the rapid expansion of algorithmic trading desks during the past few years. TABB Group, a New Yorkbased financial research company, says half of all equity trades in 2006 were done through algorithms and other electronic execution channels. Initially, algorithms were used mainly to improve and speed up order execution or to slice big orders into small transactions to mask major moves by a single trader. Using them to pick stocks based on news is a relatively new development.
“As electronic execution kept increasing, you saw more algorithms used to trade,” notes Reuters’s Brown. He says the evolution to news-based portfolios is the next natural step. “If you know what you are doing, news is pretty easy to exploit.”
At Thomson, Meagher stumbled upon the trend more than a year ago through his efforts to use computers to sort and compile financial data to help generate news stories. “Hedge funds were reverse-engineering our news and turning it back into data,” says Meagher, who discovered the practice through talks with portfolio managers.
Initial use of computer-ready news has centered on U.S. equity trading -- the area with the heaviest use of algorithms. But MIT’s Lo anticipates that incorporation of news into algorithms will quickly expand to currency trading (where political news is an important indicator) and other markets.
Hedge fund firms that are already using the technology are doing so in a variety of ways, from triggering trades to identifying trends and providing insight for investment research. In the case of HG Trading, Morgan plans to use news feeds as part of a defensive strategy -- as an early-warning system to stop programmed trades when news arrives that could change a stock’s outlook. “If my algorithm triggers a sell signal without knowing about some news, I could potentially get killed,” he says.
Even funds that don’t use news in their trading programs are keeping an eye on the development of computer-ready feeds. John Huth, director of portfolio management for $400 million Houston-based hedge fund firm Quantlab Capital Management, says he’s intrigued by the idea of access to computer-ready news archives, even though his firm doesn’t currently use news-sensitive algorithms.
Computer-ready news feeds are still far from perfect. Although some basic methods and terms are being widely adopted, bugs remain.
“Some might think, ‘The news is tagged, that is all I need,’” says Progress Software’s Palmer. “But is it tagged appropriately and consistently?” He says different standards and approaches to tagging used in early versions of the technology have created inconsistencies among providers.
Such discrepancies become more problematic as the type of news being tagged becomes more complex and as users demand more intricate tagging that not only provides the topics within the news but also identifies whether coverage is good or bad. Much of the current work being done by the writers of news-sensitive algorithms revolves around such “sentiment tagging.”
“The challenging part is trying to quantify machine- readable news -- whether it is going to move a stock, and by how much,” says HG Trading’s Morgan. “If you know that, you’ve won the world.”
The news services are frantically competing to get sentiment tagging right. Reuters launched its first attempt on April 30, using a linguistics application from Corpora, a large provider of content and document management based in the U.K., to give positive or negative ratings to news. Thomson’s Meagher says others will likely follow: “We are all heading toward the more detailed extraction -- adding some kind of sentiment indicator.”
Although the major wire services represent the front line in the development of computer-readable news, they are not the only ones that might offer it. Press release services, government sources, local newspapers and, of course, blogs are also potential news providers. How far down the news chain computer tagging goes will depend on how useful traders find the information.
Lo says there are few pieces of news that are without value to algorithmic traders. “The process of tagging is subtle, very complicated and fraught with issues,” he says. “If tagging is appropriate, it can be incredibly useful. If not, it’s just noise.”
