We are developing an algorithm that models twitter users and groups of words that may indicate real world events.

One application is modelling elections, i.e which party is likely going to win. Another application is modelling the stock market. In addition, we are also interested in how the tweets, elections and stock markets correlate with each other.

With regard to stock market, what would be sensible indicators to model? One obvious metric would be the index price movement. Another we were thinking about is volume of trade and perhaps volatility. Anyway, not really sure.

What I'm looking for is a list if indicators that would be sensible to model. What I have so far:

  • Index price
  • Specific stock prices
  • Price differences
  • Trend
  • Expected future returns

Any additions that would make sense?

  • 1
    $\begingroup$ I'm closing this question since it is way too broad and represents an entire thesis. My sincere recommendation is that you get a job in quantitative finance and work for a few years to understand what this material is. $\endgroup$ Feb 9, 2013 at 17:00

2 Answers 2


You may be interested in the twitter based hedge fund that recently fell under, but what you will be looking to model is market sentiment from the tweets and there are different ways to do this and a whole field of literature on this topic. Here is a decent thesis from MIT on the topic

  • $\begingroup$ The OP is well aware of social media and algorithmic trading. He has a history of posting broad nonsense on Quant.SE with such gems as "what education is required for HFT?" and "what are the pros and cons of C++?". $\endgroup$ Feb 8, 2013 at 14:53
  • $\begingroup$ @chrisaycock I appreciate that you still actively practice cyber-bullying even on matters that happened one and a half year ago. Would it be possible that you go waste your time elsewhere and let me post a single question on this site without you instantly deleting my account? This is not your personal club, you know... $\endgroup$
    – siamii
    Feb 8, 2013 at 15:14
  • 2
    $\begingroup$ @bizso09 he is the mod of the site you know and its hes job to keep a professional standard $\endgroup$
    – pyCthon
    Feb 8, 2013 at 15:42

I implemented (purely for testing purposes, no real world application) a similar system which was based on Google Trends, where you get data on relative search query volume for a given keyword over time (weekly intervals). The important point is that you cannot directly link social media buzz to an up- or downmove. It will however give you an idea about larger-than-usual movements in any direction.

We used a list of companies we wanted to look into, build a list of related keywords (like 'Apple' and 'iPhone'), and then aggregated the search query volume over the keywords and used a slightly modified version of the ADX Indicator. We only considered up-movements to have means to recognize when the search volume would rise exceptionally. The idea behind this was that when the search volume goes up, something is happening, or going to, without knowing whether its good or bad news, so this had to be interpreted in combination with something else, in that case we used a momentum based indicator I believe.

In a simple backtest (daily data, weekly rebalancing) it worked actually fine, but the timeseries where too short, and its pretty hard to get the required data for a sensible test, as most of it is buried in the databases of google or twitter and I'm not sure whether API-Access is possible.


Not the answer you're looking for? Browse other questions tagged or ask your own question.