Bloomberg Expands Machine-Readable News Offering With Analytics

[admin]

Bloomberg adds metrics such as most read Bloomberg-only stories and "news heat" — analysis of the most-written about companies and topics.

By Ivy Schmerken

FEBRUARY 19, 2010

Bloomberg has further expanded its machine-readable news offerings with the launch of news analytics including metrics for its service, according to the company's release today.
The metrics include, Bloomberg-only readership data: The expanded service will include an analysis of the most read stories by Bloomberg terminal users, which include market participants from top global banks, investment firms and financial institutions, from its database of more-than 30,000 global news sources. This is a unique analysis of the stories Bloomberg terminal users think are most relevant in real time.

In addition, Bloomberg provides story flow " indicating the "news heat" or most written about companies and topics by analyzing more than 30,000 global news sources tracked by Bloomberg.

"These news metrics may cause a trader or investor to rethink or alter a strategy or trading approach," said Stefan Whitney of Bloomberg, in the company's release. "This valuable information, which is derived from Bloomberg's exhaustive database of news sources and its unique user base, may alert a market participant to a market-moving event. We believe our constituents will benefit from knowing about these spikes of interest in the news," added Whitney in the release.

Bloomberg's machine-readable news offering includes tracking of 20,000 economic indicators culled from government sources, press releases and websites. In addition, it tracks company announcements, rating changes and unpredictable events. Bloomberg analyzes text to extract key information—such as earnings and revenue figures from press releases in real-time — as well as to categorize text into easy-to-identify tags for topics, tickers and people. Machines can easily interpret the text, for example, to instruct an algorithm that a story is about a certain ticker with a particular relevance (i.e. IBM, at 90% relevance) or a particular topic (i.e. a credit downgrade at 100% relevance). These signals can be used to make trading decisions or set up predefined parameters and leverage Bloomberg's machine-readable text into actionable data. The feed is delivered in Bloomberg's proprietary B-Pipe format for easy integration, and is optimized for ultra-low latency, according to the company's release.

[admin]

Researchers and Wall Street firms are analyzing the avalanche of Internet content -- including news feeds, blogs and Twitter posts -- to help improve trading performance.

By Ivy Schmerken

JANUARY 21, 2010

Finding an edge on Wall Street can be worth millions of dollars. And lately, converting text-based information -- such as news feeds, blogs and Twitter posts -- into machine-readable data is a hot topic in trading circles.

With the exponential increase in Web-based content -- including RSS-based news feeds and social networking sites such as Twitter -- as well as SEC filings, there is an avalanche of unstructured textual data to analyze. "The question is, how can you use it; should you use it?" said Mani Chandy, Simon Ramo Professor of Computer Science at the California Institute of Technology (Caltech) in Pasadena, Calif. Chandy spoke on the topic of "Analyzing Unstructured Real-Time Information for Algorithmic Trading" as part of a panel discussion at the Accelerating Wall Street virtual event in late November sponsored by Wall Street & Technology. [Ed. note: The complete event, which focused on low-latency technology and managing growing data volumes, can be accessed at www.techweb.com/wallstreet-virtual.]

"The challenge is being able to produce some alpha from the news," added panelist David Leinweber, a well-known quant and author of "Nerds on Wall Street: Math, Machines and Wired Markets," as well as the founding director of the Center for Innovative Financial Technology at U.C. Berkeley. "This can be done by filtering and categorizing news and combining it with other quantitative analytics."

It's not clear to what extent hedge funds' and quants' black-box strategies currently trade based on news feeds. According to a 2008 Aite Group report, only 2 percent of firms that employed electronic trading strategies leveraged unstructured data in a machine-readable format. But there have been advances in the past five years in services for analyzing news, blogs and other text-based reports, Caltech's Chandy noted. For example, major news providers Dow Jones and Thomson Reuters offer news products that archive the news and provide machine-readable feeds for use in algorithms.

While Chandy indicated that there are news services that provide structured data that firms can try for free, according to the Aite report, depending on the content required, pricing for a firm can range from $4,000 per month to $60,000 per month for low-latency premium content. Despite the hefty price tag, however, the market for unstructured real-time information is growing.

Aite estimated that global spending on unstructured data for electronic trading will jump from $75 million in 2009 to $115 million in 2010 and to more than $141 million by 2011. As a sign of the demand for integratable unstructured data, German exchange Deutsche Borse in November acquired Need to Know News, a Washington D.C.-based provider of machine-readable news for automated trading engines.

Uncovering Alpha

But gathering the unstructured news information as structured data sources is just the beginning. "The challenge is in designing systems and identifying news events by observing patterns that can be traded upon," Chandy told Accelerating Wall Street attendees.

In a research project at Caltech, Chandy explained, he is utilizing OpenCalais, a service from Thomson Reuters that creates structured data from unstructured information, to analyze the entire universe of English language blogs. The data is heterogeneous and is acquired from multiple sources, including stock tickers, blogs and news feeds, he noted.

"We observe an event by observing some type of pattern over time," said Chandy. For example, if someone were trading in energy stocks, he would look for the frequency of key words such as "weather" and "electricity." Another way to detect patterns in news events is to know what's normal and then to look for an anomaly, Chandy added. If the number of blogs about a company is 50 percent higher in the past day than over the past year, that would deviate from the norm and qualify as an observable event, he commented.

A system that examines blogs and Twitter posts is part of a larger, smart platform, known as a "sense and response system," said Chandy. At the center of a typical system, he related, is an enterprise service bus that routes data from multiple sources, such as OpenCalais; a source for acquiring blogs; a Reuters Data Feed (RDF) parser; a server for metadata retrieval; a server for time series analysis of stock tickers; and an event engine that determines what action to take.

"The structure is well understood," contended Chandy, coauthor of "Event Processing: Designing IT Systems for Agile Companies." "The question is, how does the end structure apply to your trading desk?"

For many desks, the use of unstructured data could improve alpha generation, Chandy added. But, "Before you commit to using a system that uses natural language text, you ought to spend some time looking at the costs," he cautioned. To determine if unstructured data adds value to the trading process, Chandy suggested firms examine five metrics: relevance, effort, accuracy, completeness and timeliness, or REACT. "If traders get a lot of irrelevant data, they will stop paying attention," he said.

Noting that advances in the news business allow reporters and traders to gain faster access to pertinent information, such as economic indicators and electronic SEC filings, U.C. Berkeley's Leinweber stressed, "The science of finding alpha in news is to define a series of news analytics." Pointing to current research he conducted against the broad S&P 1,500 universe of stocks using Thomson Reuters' NewsScope archive, which provides sentiment analysis to identify if a story about a company is positive, neutral or negative, he added, "News selection skills allow managers to profit, but you have to move beyond 'buy on the green' and 'sell on the red.' "

Leinweber explained to attendees that he spent a great deal of time refining event filters and leveraged Spotfire, a visualization tool from Tibco, a Thomson Reuters company, to create an interactive, event study explorer. Simple filters, he related, could look at the intensity of news events by counting the number of unique stories and alerts on a company. More sophisticated measures, he said, would take relevance into account, such as whether the company was mentioned in the headline only, in first sentence or later in the article body.

According to Leinweber, higher thresholds for news intensity and sentiment around a given company -- requiring two or three stories per day rather than one, for example -- produced higher returns over the index for the selected stock. "The higher returns are what we see when we use more stringent filters, a good indicator of signal quality," Leinweber told WS&T in a follow-up e-mail.

Garbage In, Garbage Out

But both Leinweber and Caltech's Chandy warned the audience to weigh the value of sources, indicating that blogs are not always written by reputable sources and are often lagging indicators of stock price movements (though they can be effective for spotting volatility in a stock). "Ninety-nine percent of blogs are irrelevant for trading or a business," Chandy asserted. "Rather than follow all Tweets [from Twitter], it may make sense to identify which sites are relevant for a given need."

News stories also can be inaccurate, the speakers cautioned. Chandy cited the case of a story about United Airlines' 2002 bankruptcy filing resurfacing on Google in 2008; the story then found its way onto Bloomberg. News-driven algorithms swung into action, driving United Airlines' stock price down to $3 per share, from its open of $12.50, before the market realized it was old news. Leinweber pointed out that social media also also introduce some possibility of manipulation, though he noted that the SEC is working hard to weed out this type of activity.

Despite the challenges of analyzing unstructured information, however, both Chandy and Leinweber insisted that there is an opportunity to generate alpha from mining this information. "This is a really exciting opportunity partly because of the volume of unstructured data and the range of people producing unstructured data, from the trusted ones to the manipulators," said Chandy. "If you can decide or determine what's really trustworthy and what's likely to make an impact, and do that quickly, that is a huge opportunity."

Added Leinweber, "Alpha comes from information and innovation. Clearly the great tide of new information is unstructured textual information, and that's a very profitable place to look for the future."
An on-demand version of Wall Street and Technology's Accelerating Wall Street virtual event can be accessed at www.techweb.com/wallstreet-virtual.