COMPANY financial reports don鈥檛 usually make for thrilling reading, but with the ability to make or break fortunes, they come under intense scrutiny. Now software that can extract information from the nuanced language of such reports could provide investors with the edge they need to stay ahead of the competition.
鈥淔inancial statements carry important information about the health of reporting companies,鈥� says at National Chengchi University in Taipei, Taiwan. But companies habitually downplay negative aspects by using ambiguous language and burying nuggets of information in pages of droning prose.
Text-mining techniques generally concentrate on single words: counting the number of negative or positive words in a body of text can give an indication of the overall tone, for example. But it is impossible to say whether certain words taken in isolation 鈥� such as 鈥渋ncreased鈥� 鈥� are positive or negative, says team member Yuan-Chen Chang. So the team designed an algorithm to recognise meaningful phrases instead ().
Advertisement
To do this, Liu and his colleagues use statistical models to automatically identify what they call opinion patterns 鈥� subjective phrases paired with an opinion holder. For example, the sentence 鈥淭he Company believes the profits could be adversely affected鈥� contains the opinion holder 鈥淭he Company鈥� and the subjective phrases 鈥渂elieves鈥� and 鈥渃ould be adversely affected鈥�.
鈥淐omputer linguistics and automated textual-information processing are one of the new frontiers in the world of finance,鈥� says of the Sauder School of Business at the University of British Columbia in Vancouver, Canada. 鈥淭his technique adds another tool to our statistical toolbox of text-mining algorithms.鈥�
Trading algorithms mostly rely on quantitative information, says Liu, 鈥渂ut it is obvious that textual information should be considered as well鈥�. For example, the team鈥檚 software could flag up phrases that don鈥檛 appear to tally with a company鈥檚 stated earnings, prompting a financial analyst to take a closer look. 鈥淣umbers can be used to convey a picture that does not correspond to reality,鈥� says , director of financial-reporting policy at the CFA Institute in London. 鈥淭hey tend not to reveal what really keeps managers awake at night. The tone of a report is a very useful complementary piece of information.鈥�
聯The software could flag up phrases in financial reports that don鈥檛 tally with the firm鈥檚 stated earnings聰
at the University of Minnesota in Minneapolis points out that sophisticated linguistic analysis is a very hard task. The software needs to learn which words are positive, which are negative and which are neither. Phrases need to appear often enough for a statistical-learning algorithm to accurately categorise them. Multi-word phrases might not occur often enough to help. 鈥淏undles of words tend to be rare things,鈥� he says.
The whole point of the account reporting system is to release information in a way that is fair to all investors, says Frank. 鈥淏ut if you can guess correctly ahead of others, you can make a lot of money.鈥� If the team鈥檚 system provides an edge, it could prove extremely valuable.
Mining text to monitor trends and opinions in the financial world is a rapidly growing field. 鈥淪ome of the news-feed providers such as Reuters already use sentiment analysis,鈥� says Antweiler. But such technology shouldn鈥檛 be relied on for automated decisions, he warns. 鈥淭he simple truth is that text mining can be helpful, but it doesn鈥檛 replace sound judgement and common sense.鈥�
Mining for a gaming smash
The success of video games could be predicted by data mining. Christian Bauckhage at the University of Bonn, Germany, and colleagues applied pattern recognition and statistical analysis techniques to data gathered from more than 250,000 players of five blockbuster games in the months after their release. They found that the decline in frequency and time people spent playing each game fit well-known mathematical models.
If similar monitoring was done during pre-release consumer testing, game publishers could use these models to predict the popularity and lifespan of a new game once it hits the market. The at the in Vienna, Austria, last month.
