Roundtable: Headline Figures vs Breaking News

As Thomson Reuters’ recent launch of its Reuters America news service demonstrates, news is still a critical input for investors. While ordinary investors are demanding more professional-level content, the face of news on the trading floor is changing, with firms looking to extrapolate trading signals from machine-readable news and event feeds. But has commoditization broken the news business? Or will news break the mold by evolving to meet the new demands of traders and algorithms hungry for more content? In this virtual roundtable, IMD assembles some of the vendors at the cutting edge of generating and processing news for today’s traders.

IMD: In today’s high-frequency world, does news still carry the same value for traders and investors as it used to? Or must it now be produced and consumed in different ways—from elementized feeds to within broader multimedia platforms—to deliver real value?
Rich Brown, global business manager for machine-readable news, Thomson Reuters: News will continue to move financial markets, but there is far greater information efficiency now. What once took days for the market to fully absorb now takes hours, and what once took hours is now seen in milliseconds. This is especially true with news about economic events such as a large payrolls miss or an unexpected interest rate decision. It is far easier for systems to react to numerical data rather than qualitative data that needs human interpretation. However, techniques such as automated news analysis can turn that qualitative information on aspects like the tone of an item—how positive, neutral, or negative it is—into a quantitative score. These scores make it easier for both machines and humans to interpret. Whether it’s for high-frequency trading or simply better context for a human to see patterns across thousands of data points, machines can enhance the value for traders and investors.

Georg Gross, head of front office data and analytics, Deutsche Börse: News will always play a vital role in trading. Change, however, is required in terms of how news is processed. High-frequency and algorithmic trading have grown rapidly over the past decade. Algo trading represents an estimated 50 to 70 percent of trading in US markets, and almost 40 percent in Europe. Automated trading is becoming mainstream in the US and Europe, and is starting to gain traction in the Asia-Pacific region.
The significant rise in algo trading demands structured news offerings that can be directly integrated into trading applications. Machine-readable formats allow for rapid and automatic processing of macroeconomic events that are relevant to algorithms. As a result, traders can bypass all of the scrolling headlines or news analysis of traditional news services and make trades based just on the economic data.
Deutsche Börse’s Market Data & Analytics division launched its “AlphaFlash” algo news feed in April 2010, delivering low-latency, machine-readable economic indicators from Europe, the US and now Asia. The service is continuously being expanded in terms of regions and access options, and we expect the demand for this kind of global news offering to further increase in 2011.

Richard Tibbetts, chief technology officer, StreamBase Systems: News is definitely important—maybe even more important then ever—but making sense of news is getting more and more difficult. It’s not sufficient to get news exclusively from one outlet—not if you want to trade on it. You need to understand what the best media channels are for the particular topic you are interested in, how fast you need to get the news in order for your trading activities to be successful, and you also need to understand the news in the context of the market.

Rob Passarella, vice president of Institutional Markets, Dow Jones: In the bits-and-bytes world of millisecond transactions, news is starting to carry even greater weight. Consider one of the key moments of the Flash Crash; stocks making massive price movements without any news events. The absence of news during price momentum was the first sign that something was wrong. Models are starting to incorporate news counts, categorization and sentiment to give a view on market activity, just as a human being would use.

Larry Rafsky, chief executive, Acquire Media: News carries more value today, both elementized and straight text. High-frequency trading is done by robots, and these robots need all the help they can get. News can now be digested and analyzed in microseconds, and signals extracted—signals that reveal both hidden alpha and hidden dangers (factors that put lie to the alpha supposedly uncovered by automated sifting of order and execution data alone).
The proof of my words is simple and grounded in the most basic of economic theory: we are now selling at much higher price points the very same news we used to sell for less. That’s incontrovertible. We have spent a lot of money reducing our latency dramatically, and traders are paying for the result.

Ryan Terpstra, chief executive, Selerity: Today, news carries an even greater value than it did before. The efficiency with which information moves through the market from new electronic channels has created an environment where breaking events can have a faster and more significant impact on the market. 
The introduction of computer algorithms into the trade initiation, execution, and risk management process has also created a greater need for new inputs that allow applications to have the same context as their human predecessors. High-frequency traders are the new market makers, so—similar to the old floor traders—automated trading applications need to stay informed of market context and react to breaking events. News remains an integral part to this process, and these systems need to recognize events and news in their own language—event data.

IMD: Have free sources of news on the internet cannibalized premium news services? What impact does this have on the quality of news? And is this contributing to other types of content, such as research, analysis and commentary services, becoming more valuable and replacing—or being used more in conjunction with—traditional news?
Tibbetts: Free sources are critical parts of today’s news ecosystem. Twitter leads the traditional news sources for some content, and RSS represents content unavailable elsewhere for certain topics. On the other hand, nothing beats solid research for market insight and understanding of economics. Often, research will help identify the key indicators for a business sector, and those can be monitored in other ways—either through traditional market data sources or through analysis of high-frequency news. Further, there are a number of non-traditional agencies that provide high-frequency feeds that aid investors who need to make quick trading decisions. Thus, although there are a plethora of news agencies out there—both free and premium—that provide a number of different quality news services, it’s really about identifying what types of news add value to your system, aggregating the news streams together, and then using the information they provide in an intelligent way.

Terpstra: In my opinion, free sources have actually increased the importance of premium news services. The vast majority of content that flows over blogs, Twitter, and other free outlets originates from reputable sources that are covering and uncovering newsworthy events. In many ways, free sources give premium services broader distribution.
Automated event-based trading has also increased the importance of recognized, trusted news sources. Markets typically don’t react when a small blogger uncovers market-moving information, but rather when a reputable source with broad distribution delivers the news. There is a lot of inaccurate information or noise in many free outlets, so using these sources in an automated strategy can present risk. Putting in place a smart, event-based strategy that consistently extracts from relevant and reliable sources is the best way to ensure proper risk management and alpha generation.
Another trend we’re seeing is stronger demand for differentiated primary commentary and research for use in pre-trade analytics. Clients are increasingly seeking differentiated insight ahead of events.

Passarella: Free news has its place, but falls short in a few key areas, especially business news. The DIY business of experts and industry people blogging makes it easier to find in-depth information about large-cap companies for research and analysis—look at how many people blog about Apple. But extensive coverage of small and mid-cap companies—which sources like Barron’s, The Wall Street Journal and Dow Jones Newswires do well for the professional—is often missing in free news. High-quality news delivered quickly is still an edge, and professional organizations do that very well.

Brown: The explosion of content sources is one of the monumental changes taking place in the news industry, and as a result, thousands of similar stories about any current event can now be found on multiple news aggregation sites. To circumvent this level of duplication, news organizations need a horizontal platform with a shared industry capability. Thomson Reuters is investing heavily in news and building such an industry platform so that journalists can focus on vertical content development and true digital innovation. Vertical journalism is differentiated by expertise, and can be based on an editorial perspective or a set of common interests. It produces the deepest and most relevant news for professionals, while true digital innovation creates meaningful customer connections and pushes the limits of interaction with news and information. In a media world moving as fast as the present one, high-value-added journalism is critical, and technology and content are completely intertwined.

Gross: Free news on the internet has certainly created challenges for legitimate financial news agencies. This is especially true for those like [Deutsche Börse subsidiaries] Market News International (MNI) and Need to Know News (NTKN), whose sole business is selling news products—compared to companies whose main business is selling financial systems or terminal products. One of the challenges is that many financial portals and “squawk” services are very active in picking up news from the original source and republishing it—often without attribution, and in most cases without any compensation to the source. Developments like the Dow Jones/Briefing.com settlement and the Theflyonthewall.com case would seem to indicate that these types of business models will have to change.
Another aspect of this evolution of the news business is that for professional users, the importance of knowing the source, credibility and speed of the news is magnified by all of the “noise” being published on the web. AlphaFlash clients know that because it is delivered by an accredited news agency, the product will be faster and more reliable than anything that comes from other types of sources. The same is true for exclusive content like a market-moving interview with a Federal Reserve official.
To the last question, it is unlikely these other types of services will become more valuable than or replace traditional news. In fact, some commentary services that have traditionally functioned largely as secondary news services will have to address the fact that there is not much added value if you’re not first with the information.

Rafsky: Other content more important than news? Of course not! Raw, breaking news is still king of the content hill. I love slow, thoughtful research, analysis, commentary, and editorial (“RACE”) as much as the next trader, but it doesn’t get the blood flowing.
And it’s not whether the news is “free”—it has to do with infrastructure spending: building your systems to react both to Reg FD disclosures (freely available) and the resulting professional commentary on corporate events from the newswires and rating firms. The public internet plays a very small role in this.
We sell news to many constituencies. Outside of trading, very little has changed. No one is dumping a paid, curated, accurate, fail-safe news service like ours for something that spiders the web. I mean, come on—one mistake like believing the zombie United Airlines bankruptcy story “broken” by Google and picked up on a Bloomberg page, and you’ve lost far more than you would ever pay me to vet the news flow for you (we recognized the out-of-date byline on that story and never sent it to any of our customers). OK, I have seen some on-line information sites dump news services like mine and not pay publishers for content, but what they lose in eyeballs that never return when a user follows a link, and how their image suffers—it’s hard to calculate damage that great. They’ll be back—and the great names have never left.

IMD: How are firms incorporating these inputs into their trading strategies—from investment managers and traders trying to get an edge, to algorithmic and high-frequency trading “black boxes”? Is news really being used to power algo trading, or more for strategy monitoring and risk management, and if so, by what kinds of firms?
Passarella: News fits across many spectrums in the investment world. Elementized news is really about extracting data points (the numbers) from unstructured data and using it to trade. That works perfectly for an algorithmic model, especially for scheduled events like earnings or economic releases in which there is a consensus estimate to benchmark against for surprise. The search for liquid trading products that correlate with indicators is the traditional venue of the high-frequency world.
The next area of focus for news is using it in a research context, such as multifactor models. Many asset managers are using news in addition to classic fundamentals for stock selection. Most of these types of models generate a longer-lived signal.
News is also used in risk management, especially when it comes to existing holdings in a portfolio and news volume. Sometimes the sheer volume of news in a sector or industry can give enough of an idea of the scale of an issue. Usually understanding these signals requires a combination of man and machine, since machines on their own still have a hard time with context.

Rafsky: News is still being used for all the things it traditionally has been—basic research, target identification, risk management, trading—algorithmic “black box” trading is just another use. Yes, it’s real. I won’t supply any details, but a certain after-market news strategy that used to make human traders money is now “owned” by the robots. There are no shares left by the time your finger hits the trade button—software beats wetware.

Terpstra: Machine-readable news and event data began as something that lived in company marketing documents and the press. Today, trading firms are absolutely using news and event data in their production automated and trading systems to generate alpha and hedge risk. The firms that we’ve seen become the most successful are proprietary and high-frequency trading firms. One of the largest and most successful proprietary trading firms in the world has built out a highly profitable, automated equity and corporate events trading desk in response to Selerity’s event data products.
Another tier-one high-frequency trading firm is now in the process of expanding its low-latency trading operations to trade corporate event data in response to Selerity’s offerings. Traditionally, this firm only used market data and economic event data in its algorithmic trading models. Selerity’s products have allowed the firm to diversify its event-driven trading strategies, and have generated significant incremental trading revenue for the firm.

Brown: There is no “one-size-fits-all” strategy when it comes to how any particular firm or class of firms will use machine readable inputs in their processes. In the early part of the lifecycle for machine-readable news, firms often used it as a circuit breaker—a defensive response—to ensure they weren’t caught off-guard on a particular piece of news. Techniques and technologies have advanced greatly over the last few years, and have enabled machine-readable inputs to be adopted in a host of use cases from quantitative trading across both high- and lower-frequency strategies to risk management, market surveillance, and even more effective circuit breakers. The increasingly robust metadata offered in these types of services can help ensure algorithms are responding to the most relevant, important, and impactful news affecting one’s portfolio. With all that in mind, we recognize most of the adoption has been in the front-office, where firms are looking to differentiate their strategies and drive increased alpha.

Tibbetts: High-frequency, alpha-seeking algorithms are not finding news to be a big source of alpha. The firms that have been successful at finding alpha in news analysis are medium-frequency traders who hold positions for days or hours.
The most common way we see news integrated into HFT systems are with news-based circuit breakers… [where] being able to leave the market or change trading strategies when extreme market events occur can prevent significant loss. So you might be looking for negative news about a company’s management team, and—rather than guessing which way the market will react—just flatten positions and avoid exposure to that event. It’s also possible to integrate the news-based circuit breaker into an execution algorithm or OMS, alerting manual traders to guide the system when unexpected news comes out. This kind of system is of interest to larger brokers looking to offer an edge to their customers. Nonetheless, there are some markets—like the commodity or foreign exchange markets—where news has a substantial impact on price and this is where you see many firms attempting to use news to trade directly, especially in the high-frequency space.

IMD: Is it truly possible to (a) build, and (b) trade on “sentiment analysis” as a reliable input? Or is “sentiment” specific to each user, and are other, more quantifiable methods of scoring news items likely to prove more useful?
Passarella: Yes, several firms have in-house models that trade using sentiment. Most of these are very specific, or situational, in nature. They are not generalist strategies. Generally, firms parse full-text feeds on their own since the “sentiment factor” is specific to their needs. But as more people and firms become exposed to this, general analysis tools are emerging. It is one of the reasons Dow Jones created Lexicon as a general sentiment feed based on coded dictionaries for the marketplace. We are starting to see the need and demand.

Brown: Automated sentiment analysis has been shown to be a valuable and statistically significant signal for many trading models. Given its systematic approach, it can be far more consistent than many other data sources over time. Sentiment, along with relevance, novelty, headline analysis and news intensity are among the key metrics scored by Thomson Reuters News Analytics, a natural language processing (NLP) system scoring some 35,000 companies and 39 commodity and energy topics. This approach is a highly comprehensive and complex method that has been in development for nearly a decade. While one can certainly try to build one’s own sentiment analytics, we encourage our clients to let us do the “heavy lifting” when it comes to the NLP work and have them focus on how to interpret it for their particular strategies. Because news and news analytics is such a rich content set, there are endless ways to explore its usage across various strategies without having to worry about the signals being diffused with increased adoption.

Tibbetts: Generic sentiment analysis is not sophisticated enough to trade on. Firms need to create and train their news analysis algorithms to specifically select the kinds of stories and information they are looking for. Simply identifying positive and negative stories is not sufficient enough for effective strategies. If your algorithm is trading on news of staff reductions, for example, then you don’t need sentiment, you need a natural language processing (NLP) system that can pick out the details of the story—like staffing counts—and understand market expectations.
Ideally, every algorithm and strategy uses customized scoring based on just the details of interest to the algorithm. That requires firms to do their own modeling and training with an open platform like the one we demonstrated with StreamBase and LingPipe [which provides language analysis software]. Everyday, improvements are being made to the body of knowledge that is “sentiment analysis,” and the early adopters will achieve the most insight into news-based trading systems and also the biggest gains. The payoffs can be substantial.

Terpstra: In our experience, we have not seen a strong uptake in news sentiment analysis for automated event-based trading. I think it’s been difficult for firms to fully understand how to systematically generate alpha from positive or negative sentiment-based signals. With that said, as the field of artificial intelligence evolves, I think using sentiment analysis as a reliable algorithmic input is possible.

Gross: Deutsche Börse already offers products that use some form of sentiment analysis. This includes MNI’s China Business Sentiment Survey, which has become a widely followed indicator of the forward momentum of the Chinese economy, and has proven to be a highly reliable predictor of official government releases. Market participants are trading on this release, and we expect to create other similar sentiment indicators in the future. While sentiment analysis has value as a trading input, the challenge is how to accurately and meaningfully quantify data that is subjective, and then disseminate it to a diverse client base. Customers are certainly hungry for these additional “trading signals,” and we are actively researching this area and working closely with them to find solutions that best fit the algorithmic news space.

Rafsky: Much of what is being peddled as “sentiment” bears no relation to what happens in the equity markets. We prefer the name “impact”—and yes, you can build it. And they will come. They have come. It works, every day, and you can trade on it. The problem is, you can’t (yet) trade much. The signal extraction is still a little too primitive. But we are getting there. It turns out that the mathematical models of market microstructure and the spread of information don’t quite fit reality as well as the finance professors believed. So we went down some blind alleys believing some specific models of volatility change before and after anticipated and unanticipated news events. The real world is a bit messy.
Now let’s talk about the emperor’s new clothes. Do you realize that some of the firms peddling volatility don’t base their measures on the statistics of the equity markets? Do you want to identify “happy” or “sad” talk, or do you want to know what moves an order book? The food industry used to use experts that told it when food tastes good. That’s like some sentiment engines. The food manufacturers that make more money use consumer panels and blind testing to determine what people will buy. Hey, it’s your choice—opinion or profit.
Trading on news works: Look at the trading in SAM after Boston Brewery gave positive guidance on Tuesday, Dec. 14. Thomson Reuters, Bloomberg, others ran with it a few minutes before the official release on MarketWire. The stock gapped up, robots made money, and people made money a little later. Heck, there was plenty of alpha left over for Wednesday, lots of time to get in on this, as long as you have your filters and alerts set right, and are paying attention. Our software, with and without impact measures, helps you do this day in and day out.

IMD: What opportunities are there for data “sources” such as exchanges and trading platforms to leverage existing content—such as company disclosures, trading and imbalance information and other datasets—to create their own news services, and how can full-service newswires and niche technology providers continue to differentiate themselves and avoid being disintermediated?
Gross: We believe there’s room for both the data sources of Exchanges, and traditional newswires to grow and adapt as this niche continues to evolve. Deutsche Börse acquired US agencies Market News International (MNI) and Need to Know News (NTKN) in 2009 as part of our overall strategy to expand our data services and offerings for the algorithmic trading community. As MNI and NTKN are both well-respected leaders in providing both traditional economic as well as machine-readable news, respectively, we created an opportunity to combine our resources and expertise to offer a best-in-class offering of algo news products.

Tibbetts: Commercial news services differentiate themselves primarily on the quality of their content, which means accuracy and insight. The market for accuracy and insight is not going away, but firms are coming to appreciate also receiving elementized, scored, actionable data from their news sources, or using technology to create their own. Exchanges—particularly in emerging economies—are learning that they, too, can create value-added data and compete for flow based on the information available on their own platforms. I don’t see newswires being completely disintermediated by exchanges, but there will be pressure on some of their reporting from that quarter.

Brown: Many of these data sources, such as exchange messages, are already in machine-readable format, but there does exist a market for other types of data such as company disclosures and other proprietary datasets. Depending on the type of information being conveyed, there may be many different use cases—even beyond the algorithmic trading market. The publishers/ creators of these datasets should work closely with the distributors to maximize the value they can attain in the marketplace. Full service newswires and the niche technology providers need to work on standards for distribution to ensure everyone is on a level playing field—from simple standards on message formats to more complicated standards on the release methods, including very stringent guidelines on the release times.

Passarella: Trading shops are used to merging data from multiple providers to forge new analysis and gain an edge. For data providers, the key is to be different. For example, [WSJ technology columnist] Walt Mossberg’s product review will move a company’s stock. Scoops and exclusives that make news also make markets. If you have an area of expertise, can map to other providers and can deliver information directly to clients’ in-house systems, you’ll do well.

IMD: What other types of developments will we see take off in the coming year? Will we see more use of “unstructured” content in trading decisions, for example—beyond macroeconomic releases and corporate earnings numbers—or more analytics such as news momentum, or tagging and linking to broader content sets?
Gross: The primary focus for algo traders who trade based on events has been on speed and access to key market-moving macroeconomic events around the world, and we expect this trend to continue in 2011. We’ve seen a greater appetite among AlphaFlash clients for additional data content and proximity hosting options in different geographic regions, particularly in the Asia-Pacific. That’s why we recently made AlphaFlash available in datacenters in Australia, Japan and Singapore, in addition to the seven datacenters in the US and Europe. This month we also started delivering economic data from China, Japan and Australia, and we will continue to explore new data content and markets that will benefit from machine-readable news offerings.

Terpstra: We continue to see firms using different types of events in their strategies and I anticipate the industry will build on the existing highly quantifiable information, such as earnings and macroeconomic data, to include less structured events. The industry has recognized that events create significant trading and risk management opportunities, and macroeconomic data and corporate earnings inputs are only the beginning. Trading on event data—such as geopolitics and credit ratings—is a game changer that will manipulate not just short-term trading strategies, but long-term portfolio approaches.

Rafsky: Unstructured news text has a lot of alpha, and we are just starting to unlock it. There is so much out there beyond economic releases and corporate announcements: There’s lawsuits, FDA and patent decisions, personnel issues, natural disasters.... In the near future, robots will be reading them all and trading, no question about that (tagging is merely a means to that end). And yes, news-linked momentum is key, as is linking to historical datasets at the time news breaks to help understand it. One thing market microstructure theory got right—once the impact of the news is absorbed, everything else is revealed in the order book and execution history.

Brown: Trading on macroeconomic releases is extremely competitive, and corporate earnings releases are quickly reaching that sensitivity as well. However, we are just beginning to exploit the signals in the mass of unstructured data available on the internet, in corporate databases or in proprietary research reports. We believe there is a significant opportunity in providing the essential context and relating concepts in the items to the larger population of data. Being able to measure not just the news momentum but the level of syndication and its effects, alerting to deviations from sentiment trends (when sentiment on Apple turns negative or sentiment on BP turns positive, for example), benchmarking the sentiment of companies versus their peers or industries and markets versus each other, and enhancing search techniques across multiple distribution and aggregation platforms are all on the horizon. Thomson Reuters is investing heavily in news and news analytics, and we’re eager to help make this information more intelligent, providing our customers with the knowledge to act.

Passarella: In the coming years, we’ll see more use of unstructured data. Since it is hard to work with, it holds a lot of promise. Getting it right means capturing gains that others cannot. Another developing area is the expansion of DMA across the globe. Algo and high-frequency trading will expand—especially in Asia—as established strategies enter new markets. Once that happens, we’ll see an explosion in the number and types of strategies employed. Combine this with the expansion of unstructured data and we may see our own Cambrian explosion in the space.