Jun 28th 2012, 12:31 by K.N.C. | LONDON
MARKETS, the classical economists instructed, rely on information. But what if there is too much of it, not too little?
The web and social-networking platforms have resulted in an explosion of words. Many firms apply artificial intelligence technology to get the gist, and use that as a trading signal.
One study in 2010 by researchers at Indiana University analyzed millions of tweets to predict the movement of the stock market three days later with an 87% accuracy. Such success has unleashed a new fashion for Wall Street quants to plug so-called "sentiment analysis" of social media into their massive models.
Until now these indicators have been fairly blunt, usually tracking a handful of companies on a two-dimensional scale of positive or negative sentiment. But on June 25th Thomson Reuters unleashed no fewer than 18,864 new indices, updated each minute. The system, developed by MarketPsych, a start-up in California, can analyze as many as 55,000 news sites and 4.5m social media sites, blogs and tweets (though on an everyday basis, the number it crunches will be much smaller). The indices quantify emotional states like optimism, gloom, joy, fear, anger—even things like innovation, litigation and conflict. And it does it across a slew of assets: 40 equity sectors, 29 currencies, 22 types of energy and materials, 12 agricultural commodities and 119 countries.
Parsing tweets to measure "innovation" or "litigation" might seem of little value, even if it can be measured accurately—a big if. The techniques of natural language processing are embryonic and highly imperfect. Tweets for example, are often ironic or sarcastic, which humans immediately understand but computers do not. However, presuming that the indices actually denote what they purport to measure, they are not so much meant for a person to use directly, but for hefty computer algorithms to factor in on a continuous basis. In that sense, relative changes over time may have merit.
This may help prevent what is known as "model crowding" or "quantagion" (a neologism of "quant" and "contagion"), explains Rich Brown of Thomson Reuters. The idea is that many funds' models rely on similar underlying data, so that when one melts-down, they all do, as happened in August 2007. And because everyone trades on mostly the same signals, the effects get exaggerated. Hence, quant investors are keen for new data sources to add to their models, to give them a unique trading strategy.
The new indices associated with an asset or country cost around $1,000 a month and go north from there. Yet are they useful? The verdict is out. Take the index that is informally called the "Bubbleometer" (pictured below), which is a measure of "speculative" conversations among investors over the web and social media platforms.
When the Bubbleometer is compared against the Standard & Poor's 500 Index between September 2009 and May 2012, it mostly follows the big swoops. But examined closely, one sees the Bubbleometer act erratically. For example, in late 2009 and early 2010 it showed speculative sentiment cooling just as prices were rising. In Spring 2010 it was first a lagging indicator that share prices would rise, then a leading indicator they would retreat. At midyear, sentiment and prices were inversely correlated. In autumn 2010 the Bubbleometer held steady within a narrow band, while the index jumped almost 15%.
Of course, this is not to say that a bit of clever maths won't uncover interesting patterns that are not visible to the human eye. It is useful measure of "bubbleocity," stresses Mr Brown of Thomson Reuters.
Successes from harnessing online sentiment analysis remain to be seen. One fund that famously began trading on Twitter signals in 2011, Derwent Capital in London, recently closed its fund (it plans to offer the metrics for free to retail investors who use its trading platform later this year). Similarly, MarketPsych, the firm that compiles Thomson Reuters' sentiment indices, formerly used the data for an in-house fund that has since been shuttered as well.
True, the value of data or an index can often be better exploited by third parties than the firm that cobbled it together—no one would think that Dow Jones would be filthy rich if it kept the industrial average to itself. However, it raises eyebrows that the firms capable of measuring market sentiment are willing to provide that data to others rather then keep it to itself. If it were so valuable, why would they make it available at all? The telling is in the using: let's see how investment performance improves or deteriorates with yet another signal on which to base trading decisions. After all, the classic economists aspired for
perfect information.
