5 ms·

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

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...

brk 17y ago
Why kid? It seems like a plausible/testable thing.
earle 17y ago
this is a sure loser based on efficient markets as this information is already priced in.
- gord 17y ago
  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.
  - borism 17y ago
    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.
djb_hackernews 17y ago
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 http://stocktwits.com I think they even have access to the real time feed.
- wildwood 17y ago
  What's more, you can even go broke with overall positive expectancy, if your position sizing strategy can't handle losing streaks.
jlees 17y ago
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!
- paraschopra 17y ago
  Hey, any chance you have raw data to process? I'd like to give it a shot (doing sentiment analysis on your data)
  - jlees 17y ago
    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. :)
gord 17y ago
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...

brk 17y ago

Why kid? It seems like a plausible/testable thing.

earle 17y ago

this is a sure loser based on efficient markets as this information is already priced in.

gord 17y ago
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.
- borism 17y ago
  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.

djb_hackernews 17y ago

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 http://stocktwits.com I think they even have access to the real time feed.

wildwood 17y ago
What's more, you can even go broke with overall positive expectancy, if your position sizing strategy can't handle losing streaks.

jlees 17y ago

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!

paraschopra 17y ago
Hey, any chance you have raw data to process? I'd like to give it a shot (doing sentiment analysis on your data)
- jlees 17y ago
  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. :)

gord 17y ago

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...