Posts

Twitter Sentiment and Volatility Index Directional Forecasting

/by

SMA has just completed a comprehensive analysis that shows the performance of classifier models, designed to predict next day directional movement for volatility indexes, improves by adding market sentiment measures derived from social media sources.  Please download the paper at:  https://socialmarketanalytics.com/research/white-papers

We present predictive models built from market data and S-Factors, a family of metrics designed to capture the signature of market sentiment as expressed in micro-blogging messages posted on Twitter. The objective of this report is to investigate the relationship between sentiment metrics generated by SMA and the volatility index of S&P 500 (VIX) and volatility indexes for individual equities (VXAPL, VXAZN, VXGS, VXGOG, and VXIBM), computed from equity option prices for AAPL, AMZN, GS, GOOG and IBM, respectively.

We used time series modelling and Logistic Regression as classifiers for predicting the direction of volatility. We tested the performance of the model with and without Sentiment Factor data. In our results, we found that the accuracy for predicting the direction of VIX using an ARIMAX-GARCH model with S-Factors was 70.86%. This was higher than the accuracy observed using a model that did not include the S-Factors (67.43%) . The same goes for most of the volatility indexes for individual equities that we picked.

Similarly, we compare the accuracy in predicting the probability of VIX going up the next day using a Logistics Regression model. The model that included S-Factors turned out to be more accurate than the model without S-Factor in all the volatility indexes for individual equities. The difference observed in accuracy was as high as almost 7.5% in the case of VXGS. The accuracy with S-factors was 62%, while without these factors it was just 54.67%.

Our analysis shows that the accuracy of a model increases by approximately 80% after adding SMA’s sentiment metrics to the model. Most of the investors are apprehensive of losses so they prefer a model that predicts the losses accurately. It is evident from our analysis that addition of S-Factors decreases the False Positive rate, thus predicting the downward movements of Volatility Indexes accurately.

Our results demonstrate enhanced predictive performance for models that include sentiment factors (S-Factors), using micro blogs like Twitter and StockTwits, as explanatory variables.

As usual, please contact us with any questions: ContactUs@SocialMarketAnalytics.com

Thanks,

Joe

/0 Comments/by

Predictive Power of SMA S-Factors

/by

Every quarter we review performance returns and statistical ratios for our family of S-Factors.   S-Score is a normalized representation of sentiment over a pre-defined look back period and is a key metric.  Below are some charts that look at the full history and YTD performance of our data across the entire universe.

Anyone can pick specific securities and instances where sentiment leads price movement; it’s a lot harder to consistently predict movements over the entire universe over a long period of time.  We pride ourselves on statistical consistency of our data over what is now 3.5 years of history.    We are the only company to track and publish these metrics, providing the most transparency.

We view S-Score >2 and S-Score <-2 as statistically significant.  An S-score of 2 means the current conversation on social media is more positive than 97 percent of prior conversations as filtered by our proprietary metrics.   When this happens the security moves higher with statistically significant consistency. The green line below represents the full history cumulative open to close return chart of stocks with a high S-Score (S-Score >2) prior to market open.  The Red line represents the full history cumulative open to close return of stocks with an extreme negative S-Score (S-Score <-2) prior to market open.  The black line represents the open to close return of stocks in the SP500.  The Sharpe and Sortino ratios for the green line (Pre-Open S-Score >2) are 1.37 and 2.23 respectively.  Sharpe and Sortino ratios for the red line (Pre-Open S-Score <-2) are -.54 and -.86. Benchmark SP500 Sharpe = .69 and Sortino = 1.08.

FullHistory

Below is the exact same chart for YTD 2015.  Sharpe and Sortino ratios show the benefit of our evolving filtering and scoring criteria.

returnYTD

SharpeYTD

Price and Tweet volume filters are commonly added when filtering stocks for sentiment.  Tweet volume represents indicative Tweet volume, once all Tweets are filtered indicative volume typically represents only 10% of the total volume of Tweets.  The below chart is the same return chart represented above with the added filter of Price day close price >5 and indicative Tweet volume > 5.  As you can see the Sharpe and Sortino ratios increase dramatically by adding simple filters.

PriceVolumeFilter

PriceFilterSort

Social media analytics is a learning process.  Our filtering and cleansing algorithms are continuously evolving.  We maintain our history as it was at each time and we keep dictionaries and accounts as a time series.

We have many more statistics employing other S-Factors and filtering criteria; please contact us for a more detailed briefing on SMA data and products.

Thanks,

Joe

/0 Comments/by