News for hackers. Stuff that matters.
Re:YCRFS 3: Things Built on Twitter by benreesman on Monday October 26, 2009 @09:12PM (#904519)
ok so here's my idea:

you take the data out of the hose and run a bunch of "machine learning" over it to figure out what twitterverse is saying about apple. or rather, apple stock.

if you can guess the opening direction the next day better than 50% of the time i'd say you would have some money on your hands.

i kid, i kid...

Re:YCRFS 3: Things Built on Twitter by jlees on Monday October 26, 2009 @11:46PM (#904757)
we're actually doing something similar to this, but have been tracking a number of stocks as well as other things.

twitter + sentiment analysis = prediction market; but in most observable cases, we're seeing twitter lag behind the market, not actually predict it. oh well!

Re:YCRFS 3: Things Built on Twitter by paraschopra on Tuesday October 27, 2009 @04:30AM (#905099)
Hey, any chance you have raw data to process? I'd like to give it a shot (doing sentiment analysis on your data)
Re:YCRFS 3: Things Built on Twitter by jlees on Tuesday October 27, 2009 @06:14PM (#906606)
Our internal data is kinda one of the things we curate and I can't really give that away, but I do have access to an academic corpus that might be redistributable. I'll have a look. :)
Re:YCRFS 3: Things Built on Twitter by djb_hackernews on Monday October 26, 2009 @10:51PM (#904685)
You don't have to be right in the market 50% of the time. In fact you aren't guaranteed to be profitable if you are right 99% of the time.

Your winners just need to be bigger than your losers.

Average winner profit * winning percent > average losing loss * loss percent.

in any case, check out http://stocktwits.com I think they even have access to the real time feed.

Re:YCRFS 3: Things Built on Twitter by wildwood on Monday October 26, 2009 @11:00PM (#904697)
What's more, you can even go broke with overall positive expectancy, if your position sizing strategy can't handle losing streaks.
Re:YCRFS 3: Things Built on Twitter by gord on Tuesday October 27, 2009 @04:41AM (#905106)
actually, dont kid.

Lets assume Information propagates according to some diffusion model, forward only in time, and after some delay affects the price of a given stock. Twitter is the perfect feed to match against high frequency stock data to measure each stocks sensitivity to particular kinds of news (location, keyword)

So you build a weighted model of what kind of news affects which stocks, and how strongly.. and do algorithmic trading on that statistical arbitrage, and exploit the market not being perfect [thus improving market efficiency].

The engine you get is quite similar to one which acts as a real-time agent looking for twitter activity on topics of interest to you...

Re:YCRFS 3: Things Built on Twitter by brk on Monday October 26, 2009 @09:33PM (#904560)
Why kid? It seems like a plausible/testable thing.
Re:YCRFS 3: Things Built on Twitter by earle on Monday October 26, 2009 @10:31PM (#904654)
this is a sure loser based on efficient markets as this information is already priced in.
Re:YCRFS 3: Things Built on Twitter by gord on Tuesday October 27, 2009 @04:44AM (#905107)
The markets aren't perfect.

But, if an algorithm can find this mispriced asset, and trade to exploit it before a human can, then that should improve the markets efficiency and reduce the arbitrage.. or so it goes.

Re:YCRFS 3: Things Built on Twitter by borism on Tuesday October 27, 2009 @09:58AM (#905407)
What is the lag of twitter compared to paid news services? Probably enough to make investing in building such system totally unattractive compared to aforementioned paid feeds.

So you'll have to make your trading decisions based on sentiment or even insider info... Very risky and not very profitable proposition...

Not that twitter is totally useless for trading, but not very useful for HFT uses you describe.