What�s Next in News Analytics: Not Social, Not Yet

News sentiment tools are rising in popularity as an additional trading indicator for hedge funds and asset managers. To maintain trust in the data, the analytics do not yet incorporate social media and microblogging sites.

Analyzing unstructured financial news for sentiment data has been no simple task. Tools have battled entity recognition, meaning understanding when "Apple" is referring to the tech company or the fruit, or when "Clinton" is referring to Bill or Hillary. Analytics has also accounted for misspellings, aliases, and abbreviations like AAPL, POTUS, or FLOTUS.

Once that hurdle is cleared, news sentiment tools aren't just capturing the "positive" and "negative" nuances of language towards entities -- analytics is helping to identify emerging relationships among tradable assets. Platforms will flag when firms, people, or brands rarely associated with one another are suddenly mentioned together. It will prompt an analysis of why that is, how frequent the mentions are becoming, and in what context. Most importantly, advances in event detection capabilities are enabling quant analysts and data scientists to add context to a story, anticipate market reaction to news, or predict creation of a business contract.

For example, analysis may find announcements of layoffs in certain conditions typically cause the market price to go up, while other conditions cause the share price to go down. Over time, an arsenal of correlations is at a client's disposal to better understand how novel the news is, and to better forecast events specific to a firm or country, or even geopolitical risk. Perhaps most quintessentially, today's systems are designed for speed, taking approximately 200 milliseconds to read, analyze, and distribute information in real-time across 1 million to 2 million news stories per day.

Truly, the capabilities of sentiment analysis have come a long way, says Armando Gonzalez, CEO of RavenPack, a real-time financial news analysis service. Gonzalez says he has seen significant increase in interest over the past three years from hedge funds and asset managers who want to add buzz, abnormalities, and sentiment to their multi-factor trading models. "Computers have a level of intelligence that supersedes people in this task. Even though it's not perfect, it's better than a subjective human."

As a trend, RavenPack has seen a growth rate of 30 to 40 percent per year for news sentiment products for the past three years, and that indicates market interest. Gonzalez adds that the product has a 90 percent renewal rate. "It continues to grow because this type of factor is becoming more accepted as a trading tool."

The only way technology like this can be successful is if the buy-side believes in the data. RavenPack and others of its ilk are running their analytics against information from trusted websites like The Wall Street Journal, PR Newswire, Bloomberg, and Zero Hedge, while staying well clear of social media and microblogging sites that lack reputation and accountability.

"I have strong views on why not to use Twitter and Facebook yet," says Gonzalez. "Customers want accountability. We can take a bunch of news from Barrons, from reporters who are investing resources to report and will be held accountable for that news, people who will be there tomorrow to back it up. With Twitter and Facebook there is no accountability, no trust, you're not able to prove the information."

He adds that there is some fantasy about Twitter's role in breaking news. Anyone can put up false information and see it become distributed in the market. And through all the noise, navigating to the original source is difficult. Computers can also easily misinterpret data. As an example, if you did a Twitter search for earthquakes in New York City it would seem they happen all the time the way people express "was that an earthquake?" or "feels like an earthquake" on social media, but if you were to check these tweets against geological records, there is no correlation.

Harnessing the good in social is the next big frontier for news sentiment, says Gonzalez, but it will be difficult to cast a net wide enough and specific enough to weed out the noise.

There is, however, some hope in concentrated platforms like StockTwits, a Twitter-like network for investors and traders. "It brings a bit more validity to the table," says Gonzalez. The community members participate, share opinions, and want people to follow them, so they have different incentives than the general population of Twitter. "We haven't yet found quantitative value in it," as the data science is not yet ready to take on StockTwits with enough validity, "but it's an exciting example of people sharing and being held accountable for opinions. I value that from a microblogging prospective."