13 comments

[ 4.9 ms ] story [ 33.4 ms ] thread
I'd be surprised if other funds aren't using this sort of technique already and just not talking about it.
I would be surprised. There's not a huge amount of trading based on text.
Is that because it has been tried and failed, or do you think there are exploitable opportunities in this area?
(comment deleted)
It's much harder to do and there's much less history. So there's a lot of effort and relatively little confidence.

If you're rentech then you can probably do this but if you're a hedge fund that's JUST doing news/twitter/etc you are probably doomed.

BTW, $25m fund is a TINY hedge fund. I've run stuff that big myself (within the context of a much larger platform.)

Also if someone brings up the Hathaway thing I shall cry.

Years of the obnoxious naming convention for twitter-based products are finally vindicated: Two and Twenty.
This number of factual errors in such a short story is amazing:

1. The total number of daily tweets is well over 100M (not 10M)

2. The 10% sampling corresponds to the gardenhose access level, and I'm seeing over 15M tweets going through "Truthy" every day.

3. Johan and Xiao-Jun are professors, not a students.

4. Johan, Huina are at IU, not Cornell

5. Xiao-Jun is in Manchester, not Cornell

Assume I have an engine that can monitor for for varied content both on the web, and on the twitter stream. I can setup rules to watch for specific content, either key-word based or source based, and then tweet that content, or post that content to some other feed.

1. How can I monetize this

2. How can I manipulate the tweet based hedge funds

If they only have 25 million pounds of assets under management then they're a "start up" hedge fund. That means that someone bankrolled them for the tune of 25% equity and a slice of the profits until they can "pay back" the 25 million pounds. I swear, "start up" hedge funds have worse financing options than normal start ups. Thank God we don't have to deal with such draconian measures.
I've noticed some serious interest from a guy at Barclays in New York to start something along these lines. He's been showing up at the NY Machine Learning Meetups talking about it. Not sure if it's for the bank, or for his own thing though.
In other news shares in the record company of Justin Bieber skyrocket after a mystery trading system bought hundreds of millions of shares.
This is an argument I've tried to put across with my site http://twittrading.com which is a pretend live stock market based on sentiment analysis of tweets.

If you could buy/sell Beiber or Gaga this system might actually work. Anything else and you've not really got enough to determine individual stocks.

If your going on the sentiment of ALL of twitter as a driver for buy/sells then maybe that might work, or maybe it will just be a useless :)