•  
• Tweet
•  
•  
• 

Parsing the News for Quants

A growing number of hedge funds are using computer-screened news as an input in their trading. What’s next in this area? The following report by Chidem Kurdas is based on a talk by Ben Goertzel, chief science officer of a firm called Quant the News LLC and director of research at the Singularity Institute for Artificial Intelligence.

He spoke at a conference last month. For more on this event, see Insider Talk.

Listening to Ben Goertzel speak about artificial intelligence and the coming “singularity” makes one think of science fiction. Nevertheless, he has some down-to-earth insights for traders. As Mr. Goertzel made clear, his talk was partly speculative but partly about developments that have already happened or are in process.
The already happened part is machine-readable news, offered by mainstream financial news services. Traders and other financial players have always tried to digest news and other information quickly. Now, computers analyze texts more rapidly than any human being can, making it possible to process and aggregate immense amounts of information.
Software classifies articles by key words, turning news stories into tradable signals. Thomson Reuters recently announced a faster version for high-frequency trading. But “automated text sentiment analysis” remains a work in progress.

Some firms offer to tailor the technology to the needs of financial users. For instance, Lexalytics software analyzes many types of texts—not just or even primarily news articles but all types of chatter including blogs, message boards, Tweets, text messages, RSS feeds. The software is designed to separate positive from negative news and remove the noise, creating sentiment scores that trading systems can utilize.

This technology is OK and works very well if there is a lot of news about something, Mr. Goertzel says. In his view, most services offered are still very simplistic and it is possible to do much more. His firm has a competing prototype technology that he describes as more accurate. Called StockMood, it uses artificial intelligence to analyze texts, classify their sentiment and measure their potential impact on a stock's price direction.

To achieve accuracy, software has to distinguish what the same keyword signifies in different contexts. Using the keyword “good” as a positive signal would result in many errors—it has different meanings in “very good”, “less good” and “not good”, for example.

Text analysis also needs to gauge the significance of fluctuations in the volume of chatter about a company or security. A number of hedge funds are doing this in various ways, said Mr. Goertzel. Increased blog and message chatter can be just a reaction to a price movement but sometimes is a precursor to price changes.

Commonsense Intelligence

Artificial intelligence is used to assess sentiment. Technology people such as entrepreneur and Singularity Institute founder Ray Kurzweil expect that AI will overtake human intelligence in a few decades. This expected event has been christened the “singularity”. It would certainly have repercussions for trading, among many areas.

A notable backer of singularity research is Peter Thiel—manager of global macro hedge fund Clarium, venture capitalist, founder of PayPal and a founding investor of Facebook. He’s even talked about how to invest in preparation for the singularity.

CHART 1

Human beings have general intelligence, in the sense that we can solve problems that did not exist when our brains evolved. By contrast, AI programs have been specialized and limited to the purpose for which they are created—the software that plays chess can’t screen message board chatter. Mr. Goertzel points out that Google is a brilliant program, but it does only one thing.

His interest is developing artificial general intelligence. As he describes it, this is an autonomous, self-reflective, self-improving, commonsensical intelligence that resembles the human mind but can expand its capabilities and become increasingly smarter than its creators. Once the singularity occurs, AGI will be able to invent technologies much faster than humans can.

The notion of a technology that is self-aware and begets better versions of itself may bring up images of HAL, the intelligent but murderous computer in the Stanley Kubrick movie “2001: A Space Odyssey”. Singularity proponents argue that the goal is to create an AGI that will be safe and beneficial—even endow it with moral principles.

Trading Systems

The implications of AGI for trading certainly sound benign. Indeed, the finance industry is seen as a potential driver of AGI research. Banks and hedge funds –as well as government agencies – are interested in more advanced programs for screening large masses of texts and data.

Mr. Goertzel’s discussion of what AGI can do for trading is relatively straightforward. He says narrow AI can’t deal with regime changes, so human intelligence has to interfere to make the transfer from one regime to another. This causes delays and losses. For trading systems to automatically adapt to major market changes, they need to be controlled by AGI.

In his view, AGI is the next stage in the evolution of trading (chart 2). It will ingest information ranging from Twitter messages to pricing data, identify shifting patterns and adapt to regime changes even if these occur frequently.

CHART 2
-------------------------------------------------------------------------------------------------------
Evolution of Trading

Human Mind → Simple Quantitative Models → Large Systems → AGI-Based Trading
----------------------------------------------------------------------------------------------------------

One could envision a system that automatically switches off the trend-following model when chatter indicates that markets will become very choppy. This smart system would switch on the appropriate trading model as market conditions change.

Paradoxically, a market that consists of intelligent machines trading with other intelligent machines would in theory become very efficient, with the result that profitable trading opportunities would vanish. But this possibility is presumably far in the future.

For now, integrating sentiment analysis of news into real-time algorithmic trading is the thing. The focus is on finding better and faster ways of using quantitative indicators derived from text to predict the future of a security’s price more accurately. The next step is probably better risk control with news analytics, which could lead to something like AGI to adjust exposures as chatter indicates changes in a trading environment.

Hopefully, there won’t be any HAL wrecking the markets!