A couple of years ago, algorithmic trading was an untapped secret with very few sell-side firms offering it to their clients. Today’s market by comparison is somewhat saturated, with sell-side firms producing a plethora of algo-based tools – almost too many for the buy-side to swallow. The latest of these is a new breed of algorithm that interprets newsflow for traders as soon as it comes off the wire.
Whispers were heard at the start of last year about this new algorithm and by mid-2006 these had grown to murmurs. Now, mid-way through 2007, those murmurs have become full song with Reuters and Dow Jones both releasing ‘tagged’ newsflow products for algorithmic traders in the past 12 months.
Traditionally, traders would wait for news updates coming through wires from providers such as Bloomberg, Dow Jones and Reuters, and as a story popped up they would quickly check through it manually (ie, using the human brain) and act accordingly upon it. This process is now being automated so algorithms can make the judgment instead, according to rules predefined by the trader. For example, when the spread between two stocks exceeds a certain amount, the algorithm triggers action to be taken after the trader has pre-programmed the algorithm to act at a known time, when a known event takes place.
Introducing tags
The original problem with this process was that the news feed was presented in plain text and algorithms had to be implemented that would read the text and try to recognise key words such as ‘profit’, ‘revenue’ and ‘growth’. Not only was it a challenge to implement algorithms that could read text in terms of pattern recognition and linguistics but it also proved to be somewhat inaccurate. Now however, news providers can ‘tag’ their feeds, that is, include elements in the news text by which the algorithm can identify specific parts of the story.
“Our previous feed, like everyone else’s, was just letters and numbers so we have set up an entirely new feed for the algorithmic trader,” explains Bob Prinsky, executive director and senior editor institutional product development at Dow Jones, who recently announced a joint venture in this space with Progress Software, an application infrastructure software supplier.
Dow Jones’ new feed has XML (Extensible Markup Language) tags attached so when an event happens – for example, a company reports its quarterly earnings or a government reports a consumer price index – there is a tag around the figures that the algorithm can read.
Journalists enter the information in a format that is essentially a spreadsheet rather than a word processor, and the computer behind the spreadsheet sends out one copy in text format and makes another copy and sends it in an elementised or ‘tagged’ format. In effect, the news has been turned into a more quantitative feed of information for all of its economic releases.
The Dow Jones offering – Dow Jones Solutions for Algorithmic and Quantitative Trading – has three parts comprising of the Dow Jones Elementized News Feed, a non-elementised News and Text Archive that has 20 years of back information, and a platform called Dow Jones News Analytics that enables traders to take past and current information to test algorithms and see how specific words have influenced the market in the past.
Where John Bates, co-founder of Apama, a division of Progress Software Corporation, believes the use of newsflow and algorithms will have particular value is in watching markets, spotting patterns and acting upon them. “All sorts of news affects the market and if a trader could interpret the news before everyone else and see how it was going to move the market then he would position himself to take advantage of that,” he says.
Time advantage
“The act of reading the number and executing the order is compressed into milliseconds, giving the grey box trader anything up to a one-second advantage over the human being clicking a mouse,” agrees Peter Green, CEO of the Kyte Group, a trading arcade and broker. “And that’s a big edge in today’s markets.”
Enabling algorithms to understand the news feed in theory allows the trader to take advantage of a price before it is potentially moved by a news story that affects the market, but some are sceptical about how much time advantage a trader gets.
| Jonathan Cohn, senior manager at Oliver Wyman, a financial services consultancy, argues that in a traditional business competing on price is a competitive differentiator but in the algo-trading space all funds have well-paid, very bright PhD-wielding employees who are all doing the same thing and are constantly leap-frogging each other. |
“I don’t want to lose sight of the fact that speed is critical,” says Mr Cohn. “But I think the understanding of the correlation between multiple factors and using the news to complement all the other data that the trader has is the main selling factor.”
The ability for the trader to use this new breed of algorithm as an effective tool also depends on the type of news event that is being looked at. Not only are certain price-sensitive news items distributed according to specific rules but it is also argued that the process is somewhat limited because expected events such as earnings reports are scheduled, traders know when they are coming out, and the way they move the markets is usually in line with what was expected.
Inconsistency issues
With so many journalists tagging the news, it is to be hoped that when traders pull the information from the news provider, whether it be Bloomberg, Reuters or Dow Jones, it is consistent so that programmers and technologists can digest the information easily. However, another limitation identified by Mr Cohn is that of the inconsistencies in tagging across different asset classes.
What one news item means to one trader may be quite different from what it means to another, he says. He suggests that a standardised language for tagging the news on different asset classes would be needed.
“The trick is to build specific and robust vocabularies that people in given industries would agree to,” he says. “Algorithmic trading is used very specifically on given desks for given strategies, they don’t pull in a whole feed against a variety of different asset types. It’s one technologist working with one desk trader and trying to design algorithms to do one thing for one asset class.”
The use of news feeds and algorithms has received a lot of press in recent months making it easy to forget that the technology is in the early stages of its lifecycle. Some do not agree that this technology is as prevalent in the industry as is being suggested. They argue that there may be select cases where quantitative hedge funds and other buy-side players are interested but not jumping on the bandwagon because the technology still has a long way to go before trading houses and banks have confidence in it.
“The only place I have seen this put effectively into use is targeting for a trader and alerting them when there is news out on a security, or industry group, or sector that he is paying attention to,” says Eric Goldberg, CEO of Portware, an automated portfolio trading software developer. “Apart from that, I have not heard anyone suggest they are actually using this.”
To what extent fund managers and banks are using news flow and algorithms is ambiguous. The buy-side community clams up when asked about their newest trading strategies but last year it was reported that the likes of Credit Suisse, JPMorgan Chase and Dresdner Kleinwort were testing newsflow capabilities for algorithms. The one certain thing that can be taken from this fuzzy picture, however, is that the ability to use news and algorithms together is a useful additional weapon for the trader to have in their armoury and finding the right mix of man and machine is necessary.
The process cannot have the instinct of a human being, says Apama’s Mr Bates. “Human instincts are something that you can’t describe with a machine and I think that’s the disadvantage of it. On the other hand, in the circumstances that you can encode, you will beat the human beings out of the gate for speed.”
Guido Hagemann, CEO of ORIMOS, a software systems provider, agrees with Mr Bates. “Humans will still be needed for a long time. It is the human that has to specify the algorithm, what it should look for, how it should work, and then humans are needed to implement the algorithm by writing the code,” he says.
The world is a long way from the apocalyptic ‘rise of the machines’ picture that some are painting as the use of algorithmic trading strategies becomes more prolific in the financial markets. It is useful to take a step back and look at what algorithms are used for in the context of algorithmic or automated trading: spotting patterns and events that indicate a trading opportunity and then acting upon it.
In this context, the algo-friendly news products that are being produced by the likes of Reuters and Dow Jones not only allow the trader to use algorithms to alert them when a stock price reacts outside pre-determined parameters as the result of a news story, but also allows traders to test their algorithmic models to see how specific news events have influenced the market in the past.
The greatest virtue, it seems, of using news feeds and algorithms together is not a question of speed or traders gaining price advantage by interpreting the news quicker than anyone else, it is more about empowering traders with knowledge and allowing them to identify patterns with algorithms on which they can determine their trading strategies in the future.
