4 ms·

Would be interesting to know how much of this reaction came from people reacting, and how much from the High Frequency apps reacting to the people reacting. I d

Would be interesting to know how much of this reaction came from people reacting, and how much from the High Frequency apps reacting to the people reacting. I don't know if they have any way of figuring that out, but it would be interesting to learn.

azmenthe 14y ago
I'm sure some HF apps made a killing :)
- ISL 14y ago
  I expect some HF apps got killed, while other HF apps got paid. Events like this and the 'flash crash' make it look like HFT is a good thing, or at least harmless. The market is durable against people doing silly things, and rewards those who do sensible things in response.
pnathan 14y ago
I am writing a tool to scrape social media and do sentiment analysis on that to provide a signal to make stock recommendations. I plan to feed it into an actual trading system at some point. So the tweet would be integrated into the system, as would any reporting of it. :-/
- RaSoJo 14y ago
  Neat. Are you planning to expose any documentation on your tool? Would love to learn how you are approaching it.
- Iterated 14y ago
  I'm a finance guy who has been reading about NLP and sentiment analysis. Any introductory books on this subject that you'd recommend for a non-linguist interested in NLP and sentiment analysis? I'm well versed in python/R.
  - pnathan 14y ago
    Sorry, I don't have any help, I'm just a regular joe looking at this stuff and taking particularly hacky ways to get the job done on the side (when I don't have other things going on. :-) ). The approach I've been taking initially is the bag-of-words approach. Looking forward I anticipate using some sort of ngram-based approach which clusters sentiment around phrases. These are stupid approaches, but I want to see how far I can push simple and stupid. ;)
- niggler 14y ago
  Warning: There was a fund that tried to do this (https://en.wikipedia.org/wiki/Derwent_Capital_Markets https://en.wikipedia.org/wiki/Derwent_Capital_Markets) and didn't exactly succeed
- chollida1 14y ago
  How would having twitter integrated into your trading system help you in this instance? My suspicion is that by the time other people start re-tweeting this the and you can get any sort of twitter sentiment determination the market will have corrected. Are you taking the other side of this and hypothesizing that you can determine twitter sentiment before the market reacts? If so please mail me, I write in house trading systems for a living and I'd love to chat.
  - pnathan 14y ago
    > How would having twitter integrated into your trading system help you in this instance? (1) In this instance, I'm pretty sure I don't know how to solve the problem naively. It went onto a theoretically legit account. I would have to cross-correlate with a couple other sources of information. At any rate, it points out to me that I need to encode a 'reliability' measure of the information. (2) Not really. I don't have the capital to operate at low-latency levels. My hypothesis is the obvious one: markets are emotional and I should be able to construct a predictor for certain industries based on emotions emitted on accounts & news snippets relating to those industries.

azmenthe 14y ago

I'm sure some HF apps made a killing :)

ISL 14y ago
I expect some HF apps got killed, while other HF apps got paid. Events like this and the 'flash crash' make it look like HFT is a good thing, or at least harmless. The market is durable against people doing silly things, and rewards those who do sensible things in response.

pnathan 14y ago

I am writing a tool to scrape social media and do sentiment analysis on that to provide a signal to make stock recommendations. I plan to feed it into an actual trading system at some point. So the tweet would be integrated into the system, as would any reporting of it. :-/

RaSoJo 14y ago
Neat. Are you planning to expose any documentation on your tool? Would love to learn how you are approaching it.
Iterated 14y ago
I'm a finance guy who has been reading about NLP and sentiment analysis. Any introductory books on this subject that you'd recommend for a non-linguist interested in NLP and sentiment analysis? I'm well versed in python/R.
- pnathan 14y ago
  Sorry, I don't have any help, I'm just a regular joe looking at this stuff and taking particularly hacky ways to get the job done on the side (when I don't have other things going on. :-) ). The approach I've been taking initially is the bag-of-words approach. Looking forward I anticipate using some sort of ngram-based approach which clusters sentiment around phrases. These are stupid approaches, but I want to see how far I can push simple and stupid. ;)
niggler 14y ago
Warning: There was a fund that tried to do this (https://en.wikipedia.org/wiki/Derwent_Capital_Markets https://en.wikipedia.org/wiki/Derwent_Capital_Markets) and didn't exactly succeed
chollida1 14y ago
How would having twitter integrated into your trading system help you in this instance? My suspicion is that by the time other people start re-tweeting this the and you can get any sort of twitter sentiment determination the market will have corrected. Are you taking the other side of this and hypothesizing that you can determine twitter sentiment before the market reacts? If so please mail me, I write in house trading systems for a living and I'd love to chat.
- pnathan 14y ago
  > How would having twitter integrated into your trading system help you in this instance? (1) In this instance, I'm pretty sure I don't know how to solve the problem naively. It went onto a theoretically legit account. I would have to cross-correlate with a couple other sources of information. At any rate, it points out to me that I need to encode a 'reliability' measure of the information. (2) Not really. I don't have the capital to operate at low-latency levels. My hypothesis is the obvious one: markets are emotional and I should be able to construct a predictor for certain industries based on emotions emitted on accounts & news snippets relating to those industries.