Abe Cofnas, Senior Sentiment Analyst

Designing and building Stock portfolios to outperform benchmark indexes is the ever-present challenge facing portfolio managers. It is perhaps more challenging in the current era where market emotions and their underlying sentiments are generating increased turbulence in the analytical landscape. Markets are constantly impacted by frequent injections of social media, and new sources of investor sentiment and chatter. Marshall Mcluhan’s prediction of the world becoming a Global Village, where “the medium is the message” has come true. Global markets are flooded by an enormous deluge of data and noise. Separating the signal from the noise out this turbulence is a major challenge for Asset Managers. A good place to begin is with sentiment data.

We know that sentiment is an important and often leading factor in price movements. The latest Consumer and Purchasing Manager surveys of sentiment always receive great market scrutiny. However, they reflect severe limitations in the state-of the art of measuring sentiment. Most sentiment analysis using the survey methodology remains at a macro-level. Expectation data are focused on economic factors such as inflation, employment, and related measures of about future economic performance. For portfolio managers, in order for sentiment to be a factor in either stock selection of portfolio construction, it needs to be much more precisely measured and predictive. What is needed is Micro-sentiment focused at the individual firm level.

Actually, measuring sentiment about a particular company is not easy to do. Sifting through thousands of documents and internet mentions of stocks is beyond a person’s ability. Such a task is an ideal application of Artificial Intelligence. Consider how one of the founders of the Artificial Intelligence field. Marvin Minsky, defined Artificial Intelligence: “The science of making machines do things that would require intelligence if done by men”. By using Artificial Intelligence and Machine Learning (AI/ML), this generation of Asset managers have the ability to actually quantify sentiment.

Let’s define sentiment and where it comes from. For asset managers, the sentiment source is expressed in Financial News documents. In contrast to sentiment that is expressed in social media, Financial News sentiment sources include SEC filings (e.g., 10-Q, filings), news articles and opinions. Using Natural Language Processing {NLP), Each document is examined for “snippets” of text that have a positive or negative emotion about an individual stock. Each selection of text is given a “sentiment score”. Through machine learning, the scoring has achieved increased validity. In other words, if a document is scored as bullish, it is in fact bullish, and a group of experts would agree with that finding. This ability to accurately score the sentiment expressed in a text is known as constructive validity. Using Machine learning the score becomes more accurate as millions of documents become part of the data mining of the algorithm.

Sentiment Portfolio Research Results.

Let’s not ignore the importance of predictive success in using sentiment analysis. The purpose of detecting sentiment in financial news is to improve the Portfolio returns, without increasing total risk. The good news is that progress has been made in recent years in getting performance enhancement of portfolios using sentiment analysis as a factor in stock selection.

Carnegie Mellon researchers studied the sentiment derived via Artificial Intelligence, Neuro Linguistic programming, and Machine learning. The study concluded: “We find a statistically significant difference in the excess alpha generated for small market cap companies when building a portfolio based on daily sentiment.” (Source: Leveraging Financial News Sentiment to Generate Alpha, LIANG, RUCHTI, KEKRE, July 2020). Cirrus Research an equity research firm has tested and chosen News Sentiment as a portfolio factor for their US Capital Appreciation Fund! This is a huge step forward, building upon the current conventional practice among Portfolio managers which is to use traditional fundamental factors like growth, quality, value and momentum. Originally, Fama and French developed a 3-factor model, which was actually a successor to the one factor CAPM model that was previously in vogue. But in 2014 they expanded the model to 5 factors. While the Fama-French 5 factor model is commonly accepted as the standard basis for understanding market returns, firms have refined these factors for constructing portfolios to outperform benchmarks. They are looking to alternative data like sentiment to find an edge in their investment strategies.

Examples of Enhanced Portfolio Returns Using Sentiment.

Let’s look at several examples of the performance of portfolios developed by StockSnips and their partners using the AI based sentiment factor.

  1. US Large Cap Sentiment Portfolio.The investment strategy and goal are to deliver higher returns than the S & P 500 at a lower volatility on a consistent basis. The trading strategy is Equal weight and the top 30 stocks in this universe are picked using a bi-factor model that blends Daily Sentiment and Sentiment Momentum. The rebalancing is weekly and the portfolio ensures that sector / industry overweighting does not occur. The portfolio also uses a hedging strategy to guard against major drawdown in the stock market.
  2. US All Cap Sentiment Portfolio.The investment objective of this portfolio is to provide long-term capital appreciation. The portfolio seeks to select US Large, Mid and Small Cap equities that are intrinsically high in News Media Sentiment quality factor and have a favorable Sentiment Momentum signal. The portfolio is actively managed with the goal of selecting the best investment opportunities leveraging sentiment trends to optimize returns.
  3. US All Cap Fundamentals and Sentiment Blended Portfolio.The portfolio seeks to select US Large and Mid-Cap equities that are accompanied by accelerating growth, rising earnings expectations, constructive valuations and favorable StockSnips Sentiment signal. The portfolio is actively managed with the goal of selecting the best investment opportunities while limiting volatility. Cirrus’ quantitative modeling identifies stocks that are likely to outperform based on fundamental variables such as growth, earnings expectations and valuations. Specifically, four stock selection pillars comprise the various Cirrus models, with the Aggressive Growth model allocating more exposure to Price and Business Momentum, and placing less emphasis on the Valuation and Quality pillars. These weights are adjusted based on KPIs that are monitored by Cirrus and will mitigate risks when market regimes change.

Conclusion

The state of the art of multi-factor portfolio construction is rapidly changing. Examples of AI based sentiment signals that outperform the S&P 500 benchmark by significant amounts are emerging. Recent research by Cirrus actually used artificial intelligence-based sentiment selection in their portfolio instruction and produced excess Alpha.

Current results bode well for the future use of sentiment as a factor in a multi-factor portfolio design. In short, it pays to get sentimental! The question that is top-of-mind is whether Artificial Intelligence tracking of sentiment adds to the performance of a portfolio. Does it achieve Alpha? The answer is yes.