Twitter Trends Beat Analysts In Predicting Wall Street | Hacker News Reader

Twitter Trends Beat Analysts In Predicting Wall Street

escapistmagazine.com

19 pointsColinWright9 comments

Following its first month of public trading, a new Derwent fund based on Twitter reported 1.85 percent return on investment. During this same time period, the S&P 500 index fell 2.2 percent, "and the average hedge fund made only 0.76 per cent."

An algorithm that outperforms the stock market for one month is not news. If they have an algorithm that can do it consistently over decades I'll be impressed.

Many funds have been trying to mine this sort of data, and presumably we don't hear about it when the results aren't so good, so this is just cherrypicking.

I can train ten chimps to pick stocks, and the best-performing of them over the first month will no doubt outperform the market significantly. But if I try to sell you my stock-picking chimp then beware.

Indeed, seems like the classic sharpshooter fallacy. Given the variability in the market there will always generally be some random betting strategy that's better than everything else within a short time period.

Here's an extensive takedown of the paper; it has lots of flaws: http://blog.someben.com/2011/05/sour-grapes-seven-reasons-wh...

Depending on the frequency that trades are placed, there are statistical methods to determine "skill" much sooner than decades.

I'm sure that plenty of hedge funds were using these feeds (blogs, twitter, etc.) for years.

They just kept their mouth shut because it's foolish to broadcast that you have an edge on your competitors... I bet that since it's out in the open, it's now obsolete...

I heard about a startup which was doing this a couple of years ago via the VC who had invested in it. It's no big secret.

I heard about that on a forum a while ago, but my point was that when you find a novel way to profit from the stock market you keep your mouth shut.

This reminds me of the movie Limitless, where the protagonist devised computer algorithms that would scan forums and other social media sites to get people's sentiments of what would happen to the stock, then successfully made 40x+ returns within months.

Small sample size.

Reply on news.ycombinator.com