Posts

Since our founding in 2012 we have been acquiring out-of-sample history to use in development of factor models.  We now have five years of history, with this history we can create statistically significant signals with monthly and quarterly holding periods.

This blog will explore the use of sentiment data for monthly and quarterly holding periods.  A trading system is explored by looking at exhaustion in social media measured by acceleration and velocity of sentiment as an indicator of price movements.  My next blog will look at traditional monthly and quarterly holding period performance based on a multi-factor sentiment model.

Acceleration and Velocity Trading System

Acceleration and velocity metrics can identify shifting sentiment.   To build these metrics, we created a new 50 period signal in addition to our traditional 20 day signal.  Raw-S is the net sentiment over a 24-hour look back period from a point of observation, derived from Tweets captured during the look back period. By aggregating over 50 periods, we create a 50-period Raw-S we call R50 as follows,

v1

For the purpose of this blog, we use 3:40 PM US Eastern Time factors to create the new metrics. At 3:40 PM US Eastern, a signal is generated near Market Close but with enough time to enter trades and execute at the Close. All trades reported in this paper are assumed to be executed at the Closing price of the day.

The R50 Factor is a raw 50 period sentiment measure.  To normalize the factor, we compute a standardized measure, using the following formula,

v2

where,

S50 is the Z-Score of R50

MA50 (R50) is 50-day moving average of the R50

SD50 (R50) is 50-day moving standard deviation of R50

S50 is a way to represent the R50 raw sentiment estimate on a standard normal curve.

 The New Metrics

We want to identify when the sentiment trend changes direction.  We derive new Velocity and Acceleration metrics from the S50 factor to identify changes in long-term sentiment and the rate of change of long-term sentiment.

We define V50, the velocity, as the one period change in S50,

v3

In our research, the velocity and the rate of change of velocity are equally important in identifying the exhaustion of a sentiment trend.

We call the rate of change of velocity, the acceleration, A50, (the second derivative of the S50),

 v4

Building A Trading Strategy Using Velocity and Acceleration

We observed a mean reverting phenomenon with longer-term sentiment. A high positive peak in the S50 sentiment was typically followed by a decrease in price. Consequently, a local minimum of S50 was followed by price appreciation.

This is the foundation of the trading signals developed.

We used various portfolio sizes to test this trading strategy. Portfolios ranged from 20 highly followed stocks on Twitter to the full SMA equity universe.

Using the metrics defined above, we used the following entry and exit signals,

 

v5

Rational for using these signals is as follows:

  1. A mean reverting relation between S50 and Closing Price.
  2. We want to capture the change and rate of change rather than just the absolute level of sentiment.
  3. A V50 of 0.5 means a change in S50, of 0.5 standard deviations and A50 < 0 is deceleration, A50 > 0 is acceleration.
  4. Due to the mean reverting nature of S50, we wanted to enter when the sentiment was decelerating and had reduced by over 0.5 standard deviations in one day.
  5. We wanted to exit when the sentiment was accelerating and had already increased by 0.5 standard deviations.

 

This was a successful trading strategy across all universes selected.  Below are performance statistics for various universe sets.  20 liquid stock portfolio universe is a universe of stocks that are highly followed on Twitter.  100 liquid stock universe is a universe of 100 stocks that are highly followed on Twitter.  Top 1000 mkt. cap and universe are self-explanatory.

v6

This data is statistically significant, out of sample, and reproducible.  To learn more about the possibilities of sentiment data in your model please contact SMA at ContactUs@SocialMarketAnalytics.com.

Thanks,

Joe

 

Social media beats the mainstream media on a regular basis.  Last week social media beat the news wire in reporting the MSFT acquisition of LNKD (blog post below) and Tuesday Twitter broke SCTY being acquired by TSLA.  This information is not theoretical – it is actionable data in our feed!

Tesla Motors lit up Twitter, yesterday, when CEO, Elon Musk came out and said their cars can float on water.  Tuesday June 21, the electric car manufacturer took everyone by surprise when they announced their decision to buy the solar panel company SolarCity (SCTY) minutes after the markets closed. The first news article to mention this came out at 4:18 PM CDT. Twitter had already gotten wind of this development 8 minutes prior with a tweet from the account “TopstepTrader”.

TSLA -SCTY

The tweet from “TopstepTrader” was deemed to be credible by Social Market Analytics’ sophisticated algorithm, which separates signal from noise to create actionable intelligence. The sentiment started to move in a positive direction the very next minute. By 16:12 CDT, SMA’s subscribers received ‘S-DeltaTM’ alerts on SCTY. The PredictiveSignalTM from SMA became positive at 16:13 CDT and at 16:18, when the first news article came out, the sentiment had already reached an extremely positive level, with Tweet volume soaring high; as was the stock price. Traders who incorporated social media sentiment from SMA into their trading models were ahead of the curve, making profits as the rest of the market was just learning of the news.

SCTY

The S-Delta metric also flagged this move.  The below chart illustrates the delta values for SCTY.  Delta represents the change in S-Score over a 15 minute lookback.  Delta values of 2 or higher are huge outliers. An SMA alarm based on Delta or S-Score would have provided an alert to this breaking news.

SCTY_Delta

To find out how you can use SMA S-Factors in your investment process contact us at Info@SocialMarketAnalytics.com

 

 

Signals derived from Twitter data have typically been viewed as shorter term signals.  There are a number of reasons for this.  One reason is the lack of out of sample data to back test trading systems on.  At SMA we now have nearly four and a half years of sentiment metrics to use in the creation of longer term signals.  Long-Term is a subjective term when discussing holding periods.  For our purposes we will be looking at trading signals that generate an average holding period of one month to three months.

At SMA we do not believe that one metric provides the full tone and context of a Twitter conversation.  That is why we publish a family of metrics call S-Factors that provide a richer view of the conversation than what is available with a single metric.

With history we have been able to look at longer term metrics and changes in security prices over longer periods.  We looked at large rapid negative changes in sentiment and determined that these sentiment movements are overreactions and lead to buying opportunities. We introduce two new metrics: Velocity and Acceleration.  The universes for these back test range from 20 large Twitter followed liquid stocks to the entire equity universe.  As you can see below these strategies identify solid buying opportunities and generate healthy average profit per share.  Please contact SMA to learn more about using sentiment to generate longer holding period trading signals with sentiment data.

Below are equity curves and trading statistics net of commissions with various universes. Overall you generate much fewer trades and hold them for longer periods of time.  The 50-day S-Score chart uses the SMA S-Score, Velocity and Acceleration Metrics.  You see that the holding periods are much longer than signals typically generated by social media.  The columns represent different universe sizes.

Slide1

200 Day S-Scores returns are below.  Again, please contact SMA for more detailed information.  Slide2

To learn more please contact us at: ContactUs@SocialMarketAnalytics.com

At Social Market Analytics we specialize in sentiment for capital markets.  However, our technology can be leveraged in many areas of sentiment.  As we expand our capabilities to look at other topics I thought it would be interesting to open up the fire hose to the full Twitter universe and capture sentiment on the current conversation about candidates and issues – boy is it!   With the upcoming debate and Iowa Caucuses we are getting a lot of good stuff on the political landscape.

Donald Trump skipping the debate is bringing strong reaction.  Sentiment for Trump the candidate has dropped quickly over the last 24 hours since he announced he wasn’t participating in the debate and his ongoing feud with Fox News.

The chart below illustrates the real-time tone of the conversation and the change of sentiment over the last 24 hours (4:00pm central 1/27/2016)  for Donald Trump as calculated by our engine.  The Yellow decreasing line represents the falling sentiment for Donald Trump and the red line illustrates the conversation is overall negative over the last 24 hours.

This is early data from our expanding capability to look at other topics, I  thought our readers would find it interesting.   To learn more about our capabilities please contact Social Market Analytics at ContactUs@SocialMarketAnalytics.com.

Keep in mind this is over the last 24 hours and will definitely change so stay tuned!!!

trump sentiment

 

Thanks,

Joe

 

People ask about the persistence of SMA sentiment signals over time.   The signal length is dependent on the S-Factor used.   S-Mean for example represents a 20 day look back period and is generally used as a longer term signal.  We looked at a theoretical strategy using a universe of the Russell 1000 and S-Factors: S-Score, S-Volume and price.  It is not intended as a proposal of a trading strategy using S-Factors as a single factor in an alpha model it does illustrate the persistence of the signal over time.

We primarily looked at the signal prior to market open instead of prior days close to include the most recent social media conversation in the metrics.  If you use a sentiment score from the prior days close you don’t include overnight information such as news, foreign markets, and indication of market open…

The universe of stocks considered is the Russell 1000 representation as of 06/26/2015.  The back test period is 2011-12-01 to 2015-11-12.  There are no transaction costs/impact costs assumptions included in this analysis.

The simulation uses S-Score > 3 / S-Score < -3 AND S-Volume > 5 AND Price at Entry > $5.  An S-Score > 3 means the current conversation is more positive than 99% of conversations over the look back period (20 days).  The reason for choosing the threshold of 3 for S-Score is motivated, in part; on research results presented by Markit using the SMA S-Factors (contact us for the research).  The position holding period is Open to Close (+2), (i.e. close 2 business days from now).  The S-Factors signals used were selected at 09:10 AM (ET).

In the below chart, the Green Curve shows the results of going long on Positive Signal (S-Score positive with the thresholds used).  Similarly, the Red Curve shows the returns of the negative (S-Score <-3, negative return is good in this case). The Blue Curve is the result of going long on Positive Signal Portfolio and short the Negative Signal Portfolio each day.   Black Curve is the ONEK ETF (Russell 1000 SPDR ETF) and serves as a market reference for the back test period.  Risk adjusted performance measures (Sharpe, Sortino) and average daily returns are presented for each filter.

The results of Open to Close (+2) Strategy using |S-Score|> 3 & S-Volume > 5 & Entry Price > $5 signal at 09:10 AM are below.  To test the strength of a signal with strong threshold, we decided to hold the position for 3 days, (buying at the open when the signal met requirements of the filter and selling it at the close 2 business days later).  If a stock appears on 2 consecutive days, the trades are made in isolation, irrespective of the position that is held due to previous signal. The Positive and Long Short outperform the ONEK benchmark.  A large spread between Positive results and the benchmark is evident.

In our simulation the positive S-Score portfolio returns 23 bps per day versus the benchmark of 9 bps per day.  The long/short portfolio returns 28 bps versus the 9 bps per day for the benchmark.  Certainly significant reason for further analysis of adding sentiment signals to your portfolio.

These results illustrate that adding S-Volume and price filters, in conjunction with S-Score, yields significant positive returns performance.

Please contact us for much more detail and other back test or to explore how sentiment metrics can be used in your environment.

R1000 S-Score

RtnsTable

Social media is a new and rich source of trading ideas.  To illustrate this point, below are some recent  trading opportunities social media data presented.  In each case activity and sentiment increase prior to the actual event.  Social media is a leading indicator of stock performance and SMA is the leader in providing metrics based on social media.

Teva acquires Allergen

Teva Pharmaceutical Industries surged in pre-market trading on July 27, 2015 on news that the company will be acquiring Allergan’s (AGN) generic drug business. Before this happened, sentiment on Twitter had already become strongly positive. At 4:00 AM EDT, when the stock price was $66.00 there was significant positive sentiment on Twitter. The sentiment rapidly shifted positive. By 7:24 a.m., the stock was trading at $72.30. The stock opened at $67.80 when the sentiment was 3.92 and closed at $72.

Figure 1:  S-Score™ For TEVA Pre- and Post-Announcement.

TevaSentiment

TevaHistoricalSentiment

Historically, daily sentiment scores for TEVA fluctuated near 0 (Neutral), with low social media activity as indicated by the time series of the S-Volume™ metric.  This behavior started to change on July 26 with significant upticks in indicative Tweet volumes and sentiment levels.  On the morning of July 27th,TEVA’s S-Score™ increased sharply to a high positve level, coincident with a spike in S-Volume™ consistent with high social media activity, indicating that SMA’s processing technology had sucessfully detected the signature of positve sentiment for TEVA embedded in the Twitter data stream.  This high positive sentiment level persisted through the open on July 28th and then started to return to typical historical levels as the markets and social media fully integrated the effect of the announcement.

Rumored Announcement of Acquisition:  Twitter (TWTR)

On July 14, 2015 at 11:39 AM EDT, a rumor started spreading on Twitter about Twitter being acquired by Bloomberg.  At 11:40 AM, there was a Tweet from user ‘beckyhiu’ indicating that Bloomberg had offered $31 Billion to buy Twitter and that Twitter was considering the offer. This rumor caused the stock price to rise rapidly. A Tweet, about 30 seconds later, at 11.41 AM,  by ‘zerosum24’ confirmed that the rumor had reached Twitter and people had started talking about it. The sentiment had started rising rapidly by this time. The changes in S-ScoreTM and S-DeltaTM were significantly positive. At 11:42 AM, the sentiment was over 2, and was statistically significant.

It was soon realized this might be a hoax and that no offer was made. At 11:42 AM, ‘TurboResearch’ questioned the credibility of the buyout offer.

There had been no official statement from Bloomberg, and hence, both the sentiment and the stock price kept rising. At around 11:50 AM, a journalist from Bloomberg Tweeted that the news was a hoax and that it was not to be believed. At this point sentiment started declining as people starting tweeting negatively. The stock price dropped rapidly.  After that, there were mostly negative comments driven by the refuted rumor.  The figures below show SMA sentiment factors leading the stock price quite accurately.

TWTRSentimentPrice

Figure 2: TWTR S-ScoreTM vs. Price

TwitterVolumeSpike

Figure 3:  Intraday   S-Volume™ Chart for TWTR

Amazon (AMZN) Earnings Announcement 

Twitter sentiment can predict stock changes even after market close, as in the case of Amazon. Amazon reported earnings on July 23, 2015. While the market consensus was that the company would not beat expectations, the conversation on social media was different.

SMA data showed a sharp increase in sentiment metrics around 2:49 PM EDT. By 2:51 PM, the sentiment on Amazon was two standard deviations higher than its typical level. The stock was trading at $480.45 at this point. At market close, it traded at $482.18, higher than the price at the time when sentiment on Amazon became positive.

It was interesting to see how the stock traded after-hours once the company reported earnings. Amazon’s stock shot up more than 17% — to $568 — from its price at 3:51 PM EDT after the company reported a surprise quarterly profit. The hidden sentiment value in Twitter data predicted what “conventional” market speculators failed to predict.

AmazonTweets

AmazonEarnings

Figure 4:  Intraday S-Score™ And S-Volume™ Behavior across Amazon’s Earnings Event.

The progression of intraday S-Score™ and S-Volume™ metrics for Amazon is shown above from 1:00 PM EDT to 4:25 PM EDT.  Amazon’s sentiment remained positive throughout the day and became significant around 2:50 PM. The sentiment saw a sharp rise post the earnings announcement after market close.

We publish our own research and analysis.  We invite you to check our Research site for new updates and publications.

Thanks,

Joe