One sign of the bubble (or, perhaps the pending apocalypse) could be the fact that the so-called Twitter Hedge Fund is oversubscribed by 250%. The Derwent Absolute Return Fund sought to raise $40m, but apparently has commitments of over $100m.
This fact should not come as a huge surprise. The Derwent fund capitalizes on three hot trends:
- Social media, specifically Twitter
- Big Data
- Algorithmic trading
Yet I am convinced that we will later look back at this fund as a prime example of a frothy market and nothing more.
Perhaps the simplest way to see that this fund is selling hype more than anything else is to look at the relative size of the Twitter logo vs. their own branding on their home page.
In fact, the Twitter logo on the Derwent page is more than twice the size of the logo on Twitter's own home page. That's no accident. Clearly, they are appealing to investors who want to get "in on Twitter" whatever that may mean.
Anyone who has done any extensive text analytics can tell you how sentiment analysis remains far from an exact science.
At its best, using professionally authored content, the best statistical models for sentiment generate results in the 70-80% range. But even those are questionable. As an example, let's look at a NYT article announcing the recent news that AT&T would acquire T-Mobile.
What's the sentiment of that story?
Well, as I read it, it's largely positive for AT&T and for Deutsche Telekom (parent of T-Mobile), neutral for Verizon, negative for Sprint, positive for Wall Street investment banks and largely negative, though with some positives, for mobile consumers.Of course, a sentiment engine will assign it a somewhat meaningless score - perhaps it's 0.5777234 positive.
Now, that's with "professionally authored" content. So let's take a look at some tweets around the deal. Using Topsy (which searches the Twitter archive), I looked at tweets from March 20-21 that mention both AT&T and T-Mobile.
The results were mostly just news headlines, rewritten or retweeted, like these:
There were a few ironic or humorous comments, of course, like this one:
I'd rate that as a negative sentiment about AT&T, but not specifically about this deal. Regardless, I doubt that the sentiment engine would parse the irony well.
Beyond the difficulty of assigning sentiment to tweets, there's a much bigger issue at play. If you look at the patterns of tweets what you find is that most are reactive rather than proactive. An event occurs, the news media puts out a story, then many others RT that story. That may be an effective way to share news and information among the masses, but from an algo standpoint I'm less convinced. Traders trade the story when it first hits (or, as the axiom goes, they buy the rumor then sell the news). Twitter sentiment is likely to be a lagging indicator, at least in the real-time world of algo trading.
I understand that the investment strategies are based on research, like this paper by researchers at Indiana University, but it's not that difficult to draw correlations when backtesting with almost any set of data. I'm pretty sure that we will look back at this as one more data point for the tech bubble.