Could natural language technology spell the end for the Masters of the Universe? Star financial traders who once commanded million-dollar salaries find their skills are no longer in such demand today. Indeed, many banks are now using computer programs to trade esoteric financial instruments.
In the latest such move, Swiss bank UBS fired its head trader for credit default swap (CDS) indices and plans to replace him with computer algorithms that trade using mathematical models, according to this Bloomberg story.
Automated trading is cheaper than employing star traders and it means the bank are less exposed to the potentially catastrophic consequences of human error or fraud — in 2008 Société Générale’s rogue trader Jérôme Kerviel lost €4.9bn on unauthorized trades. But what can computers do that human traders cannot?
Most automated trading systems try to either do something faster than a human trader, or do it more consistently. The efficient market hypothesis maintains that the price of traded asset reflects all available information.
But in the real world, there is often a window of opportunity before the information is fully reflected in the price. These windows may only be open for a few milliseconds and the individual profit on each trade may be small, but thanks to high-frequency trading strategies the advantage can be multiplied many times over.
To compete in today’s information-driven markets, traders need systems capable of fast and accurate analysis of the news content.
However, manually identifying relevant newswire articles and performing human analysis on the selected news items is a difficult task due to the sheer quantity of news releases and reports that are published during trading hours.
So in recent years, a lot of research has gone into developing systems that automate the procedure of news tracking and analysis.
Typically, this is done using a variety of natural language processing (NLP) techniques to extract the relevant information from news announcements and generate trading signals based on a set of predefined rules.
First, a learning algorithm is employed to identify and classify certain types of articles – those relating to mergers and acquisitions (M&As), for example. Then the selected news texts are processed using name entity recognition and semantic analysis.
Finally, the system analyses all the collected information with the aim of producing a trading signal. In the case of M&As, a common strategy is to buy the target firm and to sell short the acquiring firm on the expectation that the latter will overpay for the acquisition.
Of course, this is also how a human trader would operate, but the difference is that by using NLP instead of humans to analyze the texts, the trading signal is generated in milliseconds rather than the minutes that a human analyst might take.
Nevertheless, there is still an area where both computers and humans struggle to be consistent and that is predicting the market impact of a piece of information.
When a company announces its earnings, the quantitative data can instantly be extracted from the news release and compared to consensus earnings estimates. But the effect that an earnings shortfall will have on a company’s share price depends on many factors, not least the subjective assessments contained in newspaper articles.
One newspaper may title the story “IBM’s earnings drop”. A second newspaper gives the story a stronger headline “IBM’s earnings plummet”.
Clearly, the news is bad and IBM’s stock price would be expected to fall as soon as the results are released. But how far?
This is where sentiment analysis comes in to play – and presumably this is the area where trading firms will seek to build a technological advantage as the rest of their business becomes increasingly automated.
News analytics is a very young discipline and today it mostly focuses on using NLP to analyse quantitative events, such as earnings and other financial information.
Geopolitical events are much harder to capture and interpret as actionable intelligence that can drive trading algorithms, but many see this as the next frontier.