Contributor since: 2012
Company: Contextuall
Our apologies. We sourced that info from the Wall Street Journal link above.
In the article you refer to, we never claimed that Bullish sentiment on Twitter was bullish for a stock. At the time of writing that article, our sample size was too small to come to any conclusions. It was written purely as an informative "list of stocks" piece.
But for what its worth, our sample size of Twitter data is growing, and our ongoing studies show that oversold + bullish twitter sentiment is actually a pretty good contrarian indicator (as your performance analysis also suggests).
All of our existing algorithms have defined holding periods, and we publish all of our backtests (several backtest links are included in this article, for example).
Hope that clears up some of the confusion.
Hey Dr. Kris,
The "bots" on Twitter are those companies that post sel-promoting links to their own sites, and load it up with tickers to show up on stocktwits / Yahoo Finance etc.
Counting these ticker mentions in any Twitter ticker volume algorithm will likely introduce a lot of noise...
Thats an excellent idea for a trading signal! We'll see if we can get our hands on data like that...
That'll be great! Sentiment analysis is really hard to do well, curious to see what you're up to.
Its tough to focus on author performance---mainly due to the fact that they never really provide holding periods for their recommendations.
Also, its almost impossible to know what the cost basis is for an article (due to the delay between submission and publishing).
But we will be looking at the impact on stock performance if an influential author, with a large number of followers, start writing about a specific stock.
For now though, we're staying focused on macro trends--specifically focusing on ETFs (too much specific risk with individual stocks)
Yep. Exactly what we're thinking. Twitter is very noisy if you're looking at ticker mentions, but it actually becomes very predictive when you look at tweets about economic conditions, like "I just lost my job" or "have a job interview tomorrow" etc.
The data is there. The trick is to get rid of the noise and focus on the stuff that matters. Very often, that comes down to separating human users from the bots.
Hi David,
Hope all is well.
We did some testing on a small sample of subscriber trends, and happy to report that we managed to build several predictive algorithms. Here's the link to our case study:
We'll also publish this study to SeekingAlpha shortly...
Yep, its based on the most recent short data releases. The next batch of short ratios should be updated again by month end...we'll be sure to post an update.
We're looking into it. Will update you in this comment feed if we find something.
Our gut says that it should be predictive of the absolute value of alpha, at the very least.
It should be noted that the company just released an important press release, updating investors on the revenue projections. The entire press release can be found here (we've updated our previous report):
Hi MWinMD, and thanks for your feedback.
First of all, this index is calculated for a very short time interval (February 2 - March 17), so we don't think the numbers are heavily impacted by the change in enabled devices.
Essentially, we took a snapshot of social media checkins for all the companies on February 2, and took another snapshot of the same universe of stocks on March.
We normalized the data based on the number of stores in our database for each company, as to make the growth in check-ins comparable. After that, we ranked the growth numbers from highest to lowest (the complete list is shown above).
Of course, this is just the tip of the iceberg. The data is there, the trick is to structure it in a way to make it useful for investment decisions.
We'll write more reports on this data over the coming weeks.
Thanks for the feedback! Good insight into ZIP. We'll look into the JWN data set to see if there's an error. These data feeds are all works in progress :)
Care to elaborate? Any feedback would be appreciated...
True. But Google Trends just looks at the search volume trends.
We took this a step further, actually performing a sentiment analysis on each individual tweet, figuring out whether it expresses a positive, negative, or neutral opinion on the ticker.
Happy to send you the raw data if you want to have a look. Just send us a message on our profile page.
Hi outcastresearcher, and thanks for your feedback!
We did apply filters to our analysis, only focusing on tweets that were sent during Friday's trading session. That may be a possible explanation for the lower number of tweets you mentioned. (Other sites tend to show the most recent batch of tweets, which may include tweets from earlier in the week).
And yes, people often retweet the same article / concept. To us, that is a very useful piece of information to analyze, since we tend to think of Twitter as an aggregator of sentiment. (The academic paper that we described above goes into detail on this topic:
But alas, we still have a lot of work to do in refining our sentiment tools. Hopefully we'll have some proper backtested data up and running over the coming weeks.
Thanks again for your thoughts!