7 ms·

Twitter Mood Predicts The Stock Market

59 points by duck 16y ago

ivancho 16y ago
Hey, why stop at passive predicting - if we set up a few boxes with a hundred thousand twitter bots pushing "calmness" updates, we can probably make the Dow keep on going up forever (or at least 87.5% of the time) . Think about all the happy 401(k)s.
- chegra 16y ago
  Cargo Cult Science :D
  - pessimizer 16y ago
    If the markets were actually science rather than the decision making of a few huge players - if the HFT algo writers take this research seriously, I bet it'd work for at least a day or two.
- pavel_lishin 16y ago
  Careful, man. People are getting sued for writing software that fiddles with the stock market. At least, if it fiddles with it without appropriate kickbacks.
robryan 16y ago
From the paper- Third, these results are strongly indicative of a predictive correlation between measurements of the public mood states from Twitter feeds, but offer no information on the causative mechanisms that may connect public mood states with DJIA values in this manner.
meatsock 16y ago
i for one will be investing heavily in tila tequila
- code_duck 16y ago
  Invest in synthesizing the next Bieber.
Robin_Message 16y ago
From the top right of page 6: To assess the statistical significance of the SOFNN achieving the above mentioned accuracy of 87.6% in predicting the up and down movement of the DJIA we calculate the odds of this result occurring by chance. The binomial distribution in- dicates that the probability of achieving exactly 87.6% correct guesses over 15 trials (20 days minus weekends) with a 50% chance of success on each single trial equals 0.32%. Taken over the entire length of our data set (February 28 to December 20, excluding weekends) we find approximately 10.9 of such 20 day periods. The odds that the mentioned probability would hold by chance for a random period of 20 days within that period is then estimated to be 1−(1−0.0032)10.9 = 0.0343 or 3.4%. The SOFNN direction accuracy is thus most likely not the result of chance nor our selecting a specifically favorable test period. I'm not sure about the result not being chance. In particular, aren't they privileging the hypothesis by using the probability of 87.6% guesses being correct? They say the result only had a 3.4% chance of happening, yet we could assign a similar probability to a wide variety of other seemingly unlikely outcomes. I'd much rather have seen the predictor tuned on a subset of the data and then tested on the rest of the data - perhaps the authors will release their predictor and we can try it on 2009. Given that twitter and the stock market are both correlated to what is happening in the world (in fact, caused by what is happening in the world,) it is not surprising that they are correlated with each other. Somehow I don't think correlation in the other direction would get the upvotes though - stock market crash predicts sadness on twitter? You don't say. In short, I don't follow all of their statistics, but I am highly dubious of their model having any predictive power.
- b0b0b0b 16y ago
  They managed to test and train on different periods of time. February 28, 2008 to November 28, 2008 is chosen as the longest possible training period while Dec 1 to Dec 19, 2008 was chosen as the test period That said, their split exhibits enormous bias: Dec 1 to Dec 19, 2008 was chosen as the test period because it was characterized by stabilization of DJIA values after considerable volatility in previous months and the absence of any unusual or signiﬁcant socio cultural events I don't understand why they would think this is okay.
danielnicollet 16y ago
Interesting hypothesis but I will wait before trusting my savings to this approach. If twitter mood can be correlated with stock market trends, I don't see how it could fare better than traditional Socioeconomics indices which measure mood in a much more controlled and accurate way. Interpreting mood trends from Twitter data assumes some very advanced semantic and linguistic analysis of messages, which even my human brain fails to understand about 25% of the time.
blhack 16y ago
Does anybody know how they got this data? Does twitter's API offer the ability to see all incoming tweets? Glancing at their API documentation seems to suggest that they turned this ability off a while ago.
- japherwocky 16y ago
  You can get a pretty big, random chunk of the public timeline with statuses/firehose or statuses/sample http://dev.twitter.com/pages/streaming_api_methods http://dev.twitter.com/pages/streaming_api_methods
- catshirt 16y ago
  I interviewed who did semantic tweets research at [some university] and said they were given an insane sample from the past x years just by making a written request to Twitter.
- Caged 16y ago
  Here is the original paper - http://arxiv.org/pdf/1010.3003v1 http://arxiv.org/pdf/1010.3003v1. The paper claims their code and data is open, but the link is broken and I was unable to find it by browsing their site...which seems to consist of mostly blank pages.
jcroberts 16y ago
I still prefer the hem line indicator. ;)
- RickHull 16y ago
  http://en.wikipedia.org/wiki/Hemline_index http://en.wikipedia.org/wiki/Hemline_index
brendano 16y ago
I rather like Eric Gilbert's "Widespread Worry and the Stock Market" on this. http://social.cs.uiuc.edu/people/gilbert/38 http://social.cs.uiuc.edu/people/gilbert/38
fleitz 16y ago
Anyone want to open a hedge fund?
binspace 16y ago
I wonder how long it takes for the observation to influence the results, thereby invalidating the observation.
steveitis 16y ago
I am very angry about this. I was damned close to going live. My code was written. My backtesting showing amazing results. Now every douchenozzle on earth will be onboard the bandwagon, and render it useless. EVERY TIME I HAVE A GOOD IDEA SOMEONE BEATS ME TO IT! WHARBLEGLARBLE!
- steveitis 16y ago
  BTW - My 'Calmness Indicator' was simply the ratio of capital letters and exclamation marks to other text. It wasn't 87 percent, but close enough. HULK SMASH! BAD SCIENTISTS! BAD!
- pavel_lishin 16y ago
  Do you have any idea what you would have done after your project went live? Because you can still beat them to that.
  - steveitis 16y ago
    Yes. I do. I was going to bet on the supershort, and superlong funds. (2-4 x average market returns) for short term periods, and double my money every few days. I was only waiting to build capital. Now I have zero time to build capital before it's leveraged away. FUDGE! I am righteously pissed off now. I had investors on the hook. now it will all be arbitraged away from me. It's just like quicksilverscreen when Rupert Murdoch litigated it away (one of the senators who voted for the DMCA was one of the lawyers defending Fox). I am ANGRY.
    - steveitis 16y ago
      I have so damned many phonecalls to make tomorrow. A lot of powerful, and wealthy people will never talk to me again because of this. Fuck. Fuck. Fuckity. Fuck.
      - ghotli 16y ago
        You may be mad, but unfortunately this is not the place for it. Let us know if it pans out.
pessimizer 16y ago
My hypothesis: This is just another indication that stock market speculation prices spend most of the time almost completely decoupled from the value of the underlying assets. Eventually there is some triggering event, lets just call it a Lehmanbrovent, when some massively leveraged party can't pay off a creditor, and the effect cascades through the market bringing it back to a real valuation. We call this state a "catastrophe." The decoupled state we refer to as a "healthy market." A good way to test it would be to see whether the accuracy of the twitter indicator varies based on types of market activities, such as: does it completely fail on crashes, but always predict run-ups?
known 16y ago
Trading != Investing

59 points by duck 16y ago

Hey, why stop at passive predicting - if we set up a few boxes with a hundred thousand twitter bots pushing "calmness" updates, we can probably make the Dow keep on going up forever (or at least 87.5% of the time) . Think about all the happy 401(k)s.

chegra 16y ago
Cargo Cult Science :D
- pessimizer 16y ago
  If the markets were actually science rather than the decision making of a few huge players - if the HFT algo writers take this research seriously, I bet it'd work for at least a day or two.
pavel_lishin 16y ago
Careful, man. People are getting sued for writing software that fiddles with the stock market. At least, if it fiddles with it without appropriate kickbacks.

robryan 16y ago

From the paper- Third, these results are strongly indicative of a predictive correlation between measurements of the public mood states from Twitter feeds, but offer no information on the causative mechanisms that may connect public mood states with DJIA values in this manner.

meatsock 16y ago

i for one will be investing heavily in tila tequila

code_duck 16y ago
Invest in synthesizing the next Bieber.

Robin_Message 16y ago

From the top right of page 6: To assess the statistical significance of the SOFNN achieving the above mentioned accuracy of 87.6% in predicting the up and down movement of the DJIA we calculate the odds of this result occurring by chance. The binomial distribution in- dicates that the probability of achieving exactly 87.6% correct guesses over 15 trials (20 days minus weekends) with a 50% chance of success on each single trial equals 0.32%. Taken over the entire length of our data set (February 28 to December 20, excluding weekends) we find approximately 10.9 of such 20 day periods. The odds that the mentioned probability would hold by chance for a random period of 20 days within that period is then estimated to be 1−(1−0.0032)10.9 = 0.0343 or 3.4%. The SOFNN direction accuracy is thus most likely not the result of chance nor our selecting a specifically favorable test period. I'm not sure about the result not being chance. In particular, aren't they privileging the hypothesis by using the probability of 87.6% guesses being correct? They say the result only had a 3.4% chance of happening, yet we could assign a similar probability to a wide variety of other seemingly unlikely outcomes. I'd much rather have seen the predictor tuned on a subset of the data and then tested on the rest of the data - perhaps the authors will release their predictor and we can try it on 2009. Given that twitter and the stock market are both correlated to what is happening in the world (in fact, caused by what is happening in the world,) it is not surprising that they are correlated with each other. Somehow I don't think correlation in the other direction would get the upvotes though - stock market crash predicts sadness on twitter? You don't say. In short, I don't follow all of their statistics, but I am highly dubious of their model having any predictive power.

b0b0b0b 16y ago
They managed to test and train on different periods of time. February 28, 2008 to November 28, 2008 is chosen as the longest possible training period while Dec 1 to Dec 19, 2008 was chosen as the test period That said, their split exhibits enormous bias: Dec 1 to Dec 19, 2008 was chosen as the test period because it was characterized by stabilization of DJIA values after considerable volatility in previous months and the absence of any unusual or signiﬁcant socio cultural events I don't understand why they would think this is okay.

danielnicollet 16y ago

Interesting hypothesis but I will wait before trusting my savings to this approach. If twitter mood can be correlated with stock market trends, I don't see how it could fare better than traditional Socioeconomics indices which measure mood in a much more controlled and accurate way. Interpreting mood trends from Twitter data assumes some very advanced semantic and linguistic analysis of messages, which even my human brain fails to understand about 25% of the time.

blhack 16y ago

Does anybody know how they got this data? Does twitter's API offer the ability to see all incoming tweets? Glancing at their API documentation seems to suggest that they turned this ability off a while ago.

japherwocky 16y ago
You can get a pretty big, random chunk of the public timeline with statuses/firehose or statuses/sample http://dev.twitter.com/pages/streaming_api_methods http://dev.twitter.com/pages/streaming_api_methods
catshirt 16y ago
I interviewed who did semantic tweets research at [some university] and said they were given an insane sample from the past x years just by making a written request to Twitter.
Caged 16y ago
Here is the original paper - http://arxiv.org/pdf/1010.3003v1 http://arxiv.org/pdf/1010.3003v1. The paper claims their code and data is open, but the link is broken and I was unable to find it by browsing their site...which seems to consist of mostly blank pages.

jcroberts 16y ago

I still prefer the hem line indicator. ;)

RickHull 16y ago
http://en.wikipedia.org/wiki/Hemline_index http://en.wikipedia.org/wiki/Hemline_index

brendano 16y ago

I rather like Eric Gilbert's "Widespread Worry and the Stock Market" on this. http://social.cs.uiuc.edu/people/gilbert/38 http://social.cs.uiuc.edu/people/gilbert/38

fleitz 16y ago

Anyone want to open a hedge fund?

binspace 16y ago

I wonder how long it takes for the observation to influence the results, thereby invalidating the observation.

steveitis 16y ago

I am very angry about this. I was damned close to going live. My code was written. My backtesting showing amazing results. Now every douchenozzle on earth will be onboard the bandwagon, and render it useless. EVERY TIME I HAVE A GOOD IDEA SOMEONE BEATS ME TO IT! WHARBLEGLARBLE!

steveitis 16y ago
BTW - My 'Calmness Indicator' was simply the ratio of capital letters and exclamation marks to other text. It wasn't 87 percent, but close enough. HULK SMASH! BAD SCIENTISTS! BAD!
pavel_lishin 16y ago
Do you have any idea what you would have done after your project went live? Because you can still beat them to that.
- steveitis 16y ago
  Yes. I do. I was going to bet on the supershort, and superlong funds. (2-4 x average market returns) for short term periods, and double my money every few days. I was only waiting to build capital. Now I have zero time to build capital before it's leveraged away. FUDGE! I am righteously pissed off now. I had investors on the hook. now it will all be arbitraged away from me. It's just like quicksilverscreen when Rupert Murdoch litigated it away (one of the senators who voted for the DMCA was one of the lawyers defending Fox). I am ANGRY.
  - steveitis 16y ago
    I have so damned many phonecalls to make tomorrow. A lot of powerful, and wealthy people will never talk to me again because of this. Fuck. Fuck. Fuckity. Fuck.
    - ghotli 16y ago
      You may be mad, but unfortunately this is not the place for it. Let us know if it pans out.

pessimizer 16y ago

My hypothesis: This is just another indication that stock market speculation prices spend most of the time almost completely decoupled from the value of the underlying assets. Eventually there is some triggering event, lets just call it a Lehmanbrovent, when some massively leveraged party can't pay off a creditor, and the effect cascades through the market bringing it back to a real valuation. We call this state a "catastrophe." The decoupled state we refer to as a "healthy market." A good way to test it would be to see whether the accuracy of the twitter indicator varies based on types of market activities, such as: does it completely fail on crashes, but always predict run-ups?

known 16y ago

Trading != Investing