Social Market Analytics secret sauce for using Twitter Sentiment for Stocks Trading
I talk to Social Market Analytics founder Joe Gits about using Twitter Sentiment for Investments and their "secret sauce" which makes their platform effective.
By Gregory Piatetsky, May 28, 2013.
Trading on Twitter sentiment is not a new idea.
It was pioneered on a serious scale in Johan Bollen et al paper Twitter mood predicts the stock market (Oct 2010) which claimed that
collective mood states derived from large-scale Twitter feeds are correlated to the value of the Dow Jones Industrial Average (DJIA) over time. ... We find an accuracy of 87.6% in predicting the daily up and down changes in the closing values of the DJIA
Although this study was soundly criticized by many professionals, a hedge fund (Derwent) was started in 2012 using the strategy based on the paper ... and was "quietly liquidated" one month after launch, despite reportedly returning 1.86% and beating the overall market.
I recently discovered Social Market Analytics (SMA) which claims that it extracts signal from Twitter noise and uses Twitter sentiment for trading.
I talked to Joe Gits, founder of SMA to find out what SMA is doing and how different its approach is.
(Full disclosure - SMA has not paid KDnuggets in any way).
Joe told me that SMA is focused on converting social media data into actionable intelligence for the financial markets, and has a crack team of several PhDs with over 100 years of experience in building analytical products for the financial markets.
SMA is collecting currently over 600 million tweets per day.
SMA started by using tools for sentiment analysis and developed number of measures, the most important of which is
S-Score™: the weighted normalized representation of our sentiment time series over a pre-defined look back period - basically how much sentiment now differs from the average sentiment in the past for the same stock.
The idea was to look at the stocks with significant deviations (+/- 2 STD) from the average S-Score, and use them for short-term trading.
This idea was reasonable, but did not produce great results.
The breakthrough occured around June 2012 when SMA found a way to restrict attention only to qualified investors, a group of people, selected algorithmically once a month, that regularly tweet about the market. The selection is done once a month to minimize scams where people start to promote a particular stock and then disappear.
In October 2012 there were about 300,000 such tweeters-investors, and as of April 2013 their number grew to 400,000.
Furthermore, SMA also measures sentiment based on Indicative Tweets, a subset of all tweets which are relevant to stocks. SMA had 18 million such tweets in April of 2013, and they grow about 10%/month.
The results of the strategy are shown in the graph below.
Note two things about this graph:
- The green line (High S-Score Strategy) outperforms S&P 500 (black line), while red line (Low S-Score Strategy) underperforms - signaling shorting opportunity.
- The strategies begin to diverge significantly around June 2012 (blue vertical line), soon after SMA started to filter the tweets to use only "qualified investors" and "indicative tweets".
Here are some examples of Indicative Tweets for Tesla ($TSLA) from May 28, first hour after the close:
- $TSLA while this rocketed $UNXL pocketed $20 swing on $DUST
- $TSLA lol it wants my $125 target
- $TSLA for the lunch hour double - love the stock - can't imagine if I held my june 65s thou!
- lol, had $TSLA june 65's at $2.6 - sold $21, now $42 - incredible to look back
- $TSLA is probably an awesome buy right here at $110.
- Here's What Sent Tesla's Shares to Record Highs t.co/6ASSzAA3o5 #Tesla $TSLA
- Tesla $TSLA stock still motoring, probably fueled at least in part by CEO Elon Musk's tweets t.co/aEja2tJJbY
- $TSLA breakout setup + volume + high short interest = perfect storm
- Sold half of my $TSLA July 100 Calls that I bought and tweeted Friday for an 81% profit. Not bad for 2 days of work #stockaction
- Talk about a rocket stock. $TSLA has soared 190% in last month and is up 269% in last year.
- One day I will wake up to the amusing sight of 10,000 $TSLA shareholders puking up their own entrails. Just not yet
- $F is only 6 times bigger than $TSLA. Ford sells 2,000,000 cars a year. People are being idiots.
- $TSLA short interest 23mm, lowest level in a year, momentum traders beware. Natural buyers largely gone
(KDnuggets does not own any Tesla stock)
Of course, sentiment is not the only investment tool.
Joe told me that a typical hedge fund uses about 7 measures, but the stock sentiment should be one of them.
Currently SMA covers about 3,100 stocks - Russell 3000, plus 100 more requested stocks, and delivers an RSS feed with stock sentiment updated every 15 minutes (soon every minute).
You can try it for yourself at