Machine Readable Newsfeeds Gain Traction Beyond HFTs (Part 1)

Investors Look At More Ways to Gain Edge

Machine readable news services, which began as ways to quickly access core economic data, have become increasingly sophisticated and are now attracting the attention of market players outside the high-frequency trading world.

According to Adam Honore, research director for Aite Group, about 35 percent of all quant firms are now using some sort of machine readable news feed.

"The first time I did a report on it, only 2 percent were using it," Honore said. "That was 2008, so there’s been pretty substantial growth."

Major news services, such as Thomson Reuters, Dow Jones and Bloomberg, have in recent years been adapting their news-mining services originally designed to capture breaking economic data and using them to take in a host of information.

Now, they are taking these revamped products and pitching them to traders who previously had little use for ultra-fast data delivery.

Honore said it is still primarily high-frequency traders that want quick access to economic indicators, but more traditional firms are using electronic news feeds for risk management and other purposes.

The basic concept is simple. Suppose an event occurs that everyone knows will bring down a stock’s price-Honore gives the example of Steve Jobs quitting Apple. If a machine can pick up on that information and trade immediately, it can reduce the risk from an adverse event.

In practice, however, trading on breaking news can be much more complicated. In the example of Jobs quitting Apple, Honore said once things settled down, Apple’s stock would likely rise, as the market realized that there are plenty of other people at the company with great ideas, too.

"I might actually want to buy more on that event, and then when the stock rises again, sell it," Honore said.

Firms might even want to slow their trading down when their computers get certain news on a company. For instance, if a large investor is planning on selling stock in a company that then announces good news, a computer picking up on that information might delay trades until after the stock comes down off its spike.

Honore noted that such a strategy runs contrary to the whole idea of high-frequency trading, yet it is one way firms are now using machine readable news.

Rich Brown, global business manager for machine readable news at Thomson Reuters, said, though his company’s platform launched in 2006, its growth has exploded since the service began offering less structured data, including an analysis of news that comes from Reuters and third parties.

"When we launched it early on, it was a niche product adopted by the most sophisticated investors," Brown said. "Over the last few years, we’ve seen a broader adoption, beyond the higher frequency black box traders to longer term investment strategies as well as the mid to back office."

Brown said the news analytic system offered by the company can determine signals that last for months or quarters, not just milliseconds. While high-frequency traders still use the service to give their computers an edge, the system is also used by traders with less fast-paced computer models, such as those engaged in statistical arbitrage.

News can move markets in a number of ways. In addition to the scheduled release of important data, such as a company’s earnings, any piece of breaking news that is important can affect a stock price. The more news there is about a company, the more volume and volatility its stock is likely to see.

However, a news analytic system can look beyond just the number of times a company’s name turns up in the media. It can see whether an article is actually about a company or if a name is merely mentioned in passing. It can also judge how substantive a news issue actually is, helping to increase the signal-to-noise ratio, according to Brown.

This type of service is even used by traders not using computer models to make decisions. "There are still a few humans among the Wall Street traders," Brown quipped.

In addition, a news analytic system can measure the sentiment or tone of an article to see whether the news is positive or negative. Clippings services pioneered numerical ratings for articles years ago, assigning different positive and negative values to various words. A computer can now score in a second the same number of articles it once took a human an hour to score.

Thomson Reuters launched its news analytic system in 2008 at a time when many clients were trying to recalibrate their models in the wake of the financial crisis. Brown said that after traders got more comfortable with their new models, they started focusing on ways of differentiating their models from those of competitors. Inputting data from news stories became one way to make a trading firm stand out.

"News has emerged as one of these differentiated value added content sets," Brown said. "The mere fact that it’s harder to work with than pricing data suggests that there is an advantage in it that you will have over your competitors."

While a news analytic system can come up with a quantitative number to assign to a company, the more advanced users are parsing words themselves and coming up with their own scores, according to Brian Rooney, core product manager for news, research, message and mobile applications for Bloomberg Professional Service.

"There’s definitely great interest still in the text, not just taking purely a numerical output from us and leaving the text behind," Rooney said. "If you can take the story and analyze it in your own way, you’ve got the opportunity at least for your own edge."

BN Direct uses Bloomberg-generated content and content generated from press releases and public filings. For an extra fee, users can add content from sources such as The New York Times and The Financial Times.

The service can analyze how many stories are being written about a company, as well as how many on-line readers a story receives. A spike in either one can be an indication that a stock is on the move.

However, it can be challenging to decide exactly what that spike means and how to act on it. Rooney points out there’s a big difference between a story about quantifiable market data and a story about a company whose chief executive officer resigned, which could be good or bad news.

In Part 2, we will examine machine readable newsfeeds further from the points of view of Dow Jones, data provider Selerity and HFT service provider Exegy.