Skip to content

Using news sentiment to drive performance gains

Svetlana Borovkova
Svetlana Borovkova
Head of Quantitative Modelling at Probability & Partners and Associate Professor at Vrije University Amsterdam

Social media and news sentiment is increasingly interesting to quantitative analysts seeking to improve investment portfolio performance. A recent white paper from Probability and Partners, and Refinitiv, analyses the application of sentiment data to four multi-factor equity strategies, and the results are promising.

  1. Application of sentiment in news and social media can add return to multi-factor equity strategies.
  2. The highest additional return results from a combination of sentiment applications.
  3. Return could be further enhanced by the addition of sentiment acceleration to sentiment change.

For more data-driven insights in your Inbox, subscribe to the Refinitiv Perspectives weekly newsletter.

News and social media sentiment data is gaining popularity in the investment community as a source of alternative data that can be used in the search for additional return. It can also be used to limit risk of investment portfolios.

Considering these potentially positive outcomes, the Refinitiv white paper Adding Sentiment to Multifactor Equity Strategies explores four applications of sentiment.

Using news sentiment to drive performance gains

Download the white paper Adding Sentiment to Multifactor Equity Strategies

Research elements

The research sources sentiment data from the Refinitiv News Analytics (RNA) engine, which uses natural language programming (NLP) algorithms to read and interpret real-time news on the Reuters newswire.

Each news item is tagged in relation to a company or commodity, and the algos provide quantitative characteristics for each item:

  • A sentiment score indicates the probability of a news item having a positive, negative or neutral impact on the price of an asset.
  • A relevance score indicates how relevant a particular news item is to the asset.
  • Novelty scores provide varied indicators, including the number of similar news items that have been identified before and after the item that is scored for sentiment.

A change in sentiment caused by news covering a stock becoming more positive and therefore driving its price, rather than a stock’s level of sentiment, is used to capture the influence of news on stock prices.

Market data used in the research is the stock prices of the S&P 500 index constituents from 18 January 2008 to 31 December 2018, the equivalent of 36 periods of 13 weeks.

The Fama & French factor model is used as the traditional investment model, with factors in the analysis including excess market return, size, value, momentum, low-volatility, and sentiment. The investment portfolio is constructed using the alpha momentum strategy from Hühn & Scholtz (2018).

Applications of sentiment

Sentiment as a separate investment factor

The research for this application of sentiment uses standard, standard plus sentiment, and sentiment-only factor models. The performance of the three associated portfolios is measured weekly on a total return’s basis.

Factor model-based investment results:

Factor model-based investment results. Using news sentiment to drive performance gains

The table above shows that the annual return of the standard factor model exceeds the return of the reinvested S&P 500 index by 2.58 percent.

A comparison of the standard factor model and the standard model plus sentiment shows an additional annual return of 0.5 percent, while risk characteristics (volatility and tracking error) are not materially different.

A comparison of a factor model that has only market return and sentiment as factors, and the standard factor model, shows a gain close to that of the standard factor model compared with the reinvested S&P 500.

Performance factor model with sentiment vs. model without sentiment:

Another angle to evaluate the added value of the standard factor model plus sentiment is to apply relative framework in which the standard factor model plays the role of the index.

The table above shows that the factor model with sentiment tracks the model without sentiment quite closely. The information ratio of 0.43, however, means 43 basis points of additional return annually over the period of nine years noted for each 1 percent of tracking error.

Clearly, sentiment as an additional factor in a factor model can add significant value.

Sentiment as a risk overlay for individual stocks

The application of sentiment as a risk overlay is based on two categories of strategies: avoiding losers and picking winners. Both of which have the potential to enhance returns, while the former can also help avoid losses.

The risk overlay is used at the stock level within sectors to avoid stocks that score lowest in terms of weekly change in sentiment. Varying percentages of stocks with the lowest sentiment change can be screened out.

In this case, three scenarios determine screening — ‘subtle’ eliminates 10 percent of stocks, ‘medium’ eliminates 30 percent, and ‘aggressive’ 50 percent.

Sentiment as a risk overlay for sectors

Just as sentiment can be used as a risk overlay for individual stocks, it can also be used as a risk overlay for sectors. Similar to the stock level example, this over-weights and under-weights sectors on the basis of sector-wide sentiment changes.

The top five sectors, in terms of sentiment, in the 11 sectors of the S&P index are over-weighted, and the bottom five under-weighted. Scenarios are again subtle; the top sectors are over-weighed 10 percent, medium 30 percent, and aggressive 50 percent.

Sentiment as risk overlay and for sector rotation:

Sentiment as risk overlay and for sector rotation. Using news sentiment to drive performance gains

The combinations of subtle, medium, and aggressive sentiment application for both the risk overlay and sector approach provide nine strategies in total. Their performance is shown in the table above.

Combinations of the above strategies

In the white paper, the final element of exploration demonstrates the effect of combining all three of the above sentiment applications.

The key result here is the level of return, the highest of all the sentiment applications and a reflection of a 2 percent gain above the index from the factor model, and a 2 percent gain from the sentiment only model.

What can we learn from social media and news sentiment?

While the application of sentiment in the investment world is still relatively new, the Refinitiv white paper highlights how it can be applied to improve portfolio performance.

It also opens the door to further research, such as the addition of sentiment acceleration, essentially, how quickly sentiment is changing for a particular stock, to sentiment change. Sentiment could also be applied to investing in other markets, such as commodities and bonds.

Download the white paper Adding Sentiment to Multifactor Equity Strategies

Refinitiv is now on Telegram! Receive daily updates of critical and timely market analysis to your mobile. Subscribe to

Subscribe to the Refinitiv Perspectives Weekly Newsletter