Using media sentiment to unearth stock market winners has big potential for traders or wealth managers. Based on 6,000 stocks over 13 years, MarketPsych research highlights the real-time influence of sentiment momentum strategies in the search for alpha.
- Sentiment momentum strategies based on thousands of news and social media sources have the potential to augment more traditional factor models.
- Studies conducted over a 13-year period by MarketPsych show the global correlation between stock performance and media sentiment values.
- Sentiment data streams such as Refinitiv MarketPsych Indices are being fed into research tools, stock screens, and heat maps in order to find new investment ideas.
“As there are so many people who cannot wait to follow the prevailing trend of opinion, I am not surprised that a small group becomes an army.” — Josef de la Vega, 1688, Confusion de las Confusiones
Josef de la Vega’s Confusion de las Confusiones was the first book written on financial markets. De la Vega observed that the transmission and spread of opinion drove price trends on the Amsterdam Exchange.
Following this early reference, numerous successful investors, including the wealthiest private individuals of their times — Japan’s Munehisa Honma and Warren Buffett of the United States — have written about the role of understanding market psychology in successful investing.
In this article, we examine a simple and consistent global investment strategy based on investor psychology called “sentiment momentum”.
The sentiment of the crowd
Using text analysis of media, we can now see (and test) the phenomenon that de la Vega wrote about over 300 years ago. Modern technologies allow us to scan and quantify the sentiment expressed in media about every asset within milliseconds.
The resultant media sentiment scores represent the general feeling about each company. Using such scores, companies can be ranked from most positive to most negative or represented in colorful heatmaps.
Figure 1 is a heatmap of top global stock sentiment from 1 September, 2018 to 1 September, 2019. Note that Xiaomi appears as one of the most positive stocks in Figure 1. According to de la Vega’s observations, investors should follow the positive trend of opinion and buy Xiaomi stock.
But is that true?
Figure 1: Heatmap depicting the most positive global stocks according to the media over the past year, where green is positive and red is negative media sentiment. Box size correlates with average daily buzz volume.
Among the global clients of sentiment data, many deploy systems like that described by de la Vega — following the sentiment of the crowd.
Wealth managers use sentiment-based stock screeners to identify future winners. Brokerages use sentiment heatmaps to help clients generate ideas. Private bankers and financial advisors recommend stocks highest on the sentiment rankings in most markets, using rankings such as that depicted in Figure 2.
Figure 2: Most positive global companies displayed with their average Refinitiv MarketPsych Indices sentiment score from 1 Jan, 2019 to 21 Aug, 2019.
Despite the extensive use of sentiment-based models by our clients, MarketPsych’s research team asked the question, “Are such models supported by evidence?”
In order to investigate this question in detail, our research team performed a series of studies on stock sentiment, using monthly and yearly aggregations of media sentiment data. They used the Refinitiv MarketPsych Indices sentiment scores, which are available in minutely, hourly, and daily frequencies for over 15,000 global stocks, 62 stock indexes, 36 commodities and 45 currencies.
- The Refinitiv MarketPsych Indices data represent aggregate scores from thousands of news and social media sources including news feeds, blogs and chat rooms.
- After pre-filtering, over two million financial articles are analyzed using a patented natural language processing technique. Refinitiv MarketPsych Indices clients in more than 20 countries use the data feed for purposes including research, alpha generation, risk management, visualizations, and market monitoring.
- For stocks, Refinitiv MarketPsych Indices covers 35 distinct sentiments including emotions such as fear, trust, and surprise and themes such as earnings forecast, price forecast, fundamentals strength, company innovation, and management change. These factors are used as the basis of research on asset prices.
The sentiment index is a time series of the net positive versus negative references to a given company. Every minute, the business and financial news and social media is scanned for new references to tens of thousands of assets.
For example, the sentence “Tesla is performing great” is scored as a positive sentiment value (+1) for Tesla (TSLA.O). When thousands of such references in millions of daily online new articles are aggregated, a net sentiment score (positive vs negative balance) can be derived and plotted over time, giving an overview of the media tone expressed about a specific firm.
Divergence in performance
MarketPsych’s Changjie Liu and Anthony Luciani performed the following studies.
They ranked stocks each month using their average Refinitiv MarketPsych Indices Sentiment values. They looked at the performance of the largest 6,000 US stocks over the period 2006-2018 based on the companies’ sentiment averages over the past month.
They basket the stocks into decile tiers and tracked their subsequent 90-day returns. The average decile performance is plotted in Figure 3.
The most negative stocks by media sentiment decile (red lines) in the prior month are also the worst price performers over the following 90 days. The 90-day forward price spread between the most positive and most negative deciles is investable.
Average subsequent price drift of stocks ranked by 30-day ‘sentiment’ Refinitiv MarketPsych Indices average (Rankings are grouped into deciles. Prices centered at day of ranking. Period: 2006-19)Figure 3: Decile returns of U.S.-domiciled stocks ranked by average sentiment over the prior 30 days and tracked forward over the subsequent 90 days. The most positive deciles (green) significantly outperform the most negative (red) over 90 days.
There is a significant, steady and consistent divergence in performance between the upper and lower deciles. The spread averages 3.5 percent annualized between the top and bottom sentiment deciles from 2006-2018. Other Refinitiv MarketPsych Indices show similar and unique predictive power.
Is sentiment momentum a winning strategy?
Sentiment momentum strategies show value outside the United States and over longer time periods. In further research, the MarketPsych team explored a one-year holding period investment strategy. In this strategy the most positive stocks — based on the prior 12-months’ media sentiment — were bought and held for one year.
The strategy started investing in January 1999 with one-twelfth monthly allocations of the initial $1 investment. Each month companies were re-ranked based on their past 12-months’ average sentiment, and the portfolio positions were rebalanced.
For the United States, the top 1,000 stocks by media buzz over the prior year comprised the selection pool, from which the most positive 200 were bought. For most other countries, except where noted, the top 100 stocks by media buzz were selected over the prior year, and a portfolio of the 20 most positive stocks was selected in each country.
In Figures 4 and 5 below, the growth of US$1 is depicted in the rising blue line while the local benchmark index is plotted in the green line.
No transaction costs or taxes are accounted for. There is significant survivorship bias in the early years for several countries, nonetheless, outperformance continues through the present.
Sentiment momentum is global: Japan, Germany, United States, United Kingdom
Figure 4: Compounded returns (blue lines) of a one-year holding period positive sentiment strategy using stocks from each domicile versus the local benchmark index (green line). Top 30 media buzz used for selection pool in Germany.
Sentiment momentum is global: Australia, Canada, China, Eurozone
Figure 5: Compounded returns (blue lines) of a one-year holding period positive sentiment strategy using stocks from each domicile versus the local benchmark index (green line).
The outperformance of high sentiment stocks is evident in every country and region we’ve tested including four not displayed — Brazil, Hong Kong, India, and South Africa.
While this model shows that highly positive stocks tend to outperform their benchmarks from a month to a year, eventually the media darlings fall from favor. As their sentiment falls, so does the stock price (often following the media sentiment lower).
Figure 6, below, depicts how falling media sentiment often appears to precede a decline in stock values. The daily value of the S&P 500 index is depicted in yellow price bars from 1 May 2019 to 21 August 2019. Two moving averages of media sentiment about the U.S. stock market — a 10-day and a 30-day average — are superimposed.
When the recent media is more positive about the stock market, there is green shading between the lines as the 10-day average rises above the 30-day average. When the recent media is more negative, red shading appears between the two averages.
Figure 6: The value S&P 500 index appears to rise and fall in correlation with media sentiment.
We’ve seen that stocks tend to outperform when the associated media sentiment is high. When that media sentiment falls, it may be a warning signal that prices are likely to decline as well.
Finding new investment ideas
With the convergence of complex natural language processing, cloud computing, and online news and social media, detailed sentiment data was developed. Modern analytic tools allow us to historically test whether buying on media sentiment is likely to lead to outperformance (or the opposite!).
There is a complex interaction between sentiment, fundamentals, and asset prices. However, based on the above research, it is not necessary to develop complex models with sentiment in order to find outperforming investments.
A simple sentiment momentum strategy shows promise in periods from monthly to yearly windows in every country studied. Being a pure sentiment factor, the model has the potential to augment more traditional factor models.
Sentiment data streams like the Refinitiv MarketPsych Indices are now being fed into research tools, stock screeners, heatmaps, and charts by our global clients.
With these tools, their users quickly find new investment ideas, better explain market price movements, and avoid stocks when the news turns negative.