Viv Penninti and Ravi Koka, Co-Founders, Stocksnips Inc
Is Stock Sentiment a Leading Indicator of Stock Price Changes?
This article discusses the potential value of using stock sentiment data generated using StockSnips’ proprietary stock sentiment algorithm. In particular we present initial findings from our study of Apple’s stock price performance versus estimated sentiment scores. We find a very high correlation between stock price and stock sentiment (correlation coefficient = 0.72; p-value <0.10). More importantly, we find that sentiment is a leading indicator of stock performance which could potentially be used for better portfolio management. While this study is limited in that only Apple stock is analyzed, it is clear that stock sentiment could be used as a key signal for stock performance.
What is Sentiment?
Stock sentiment analysis is based on the positive statements about a company. The source for assessing positive or negative sentiment could be driven from Twitter or Facebook data (Social Media or consumer sentiment) and/or it could be based on the analysis of news articles and SEC filings (News sentiment). Depending on type of news sources, it could be expected that sentiment signals from news articles would be “less noisy” and more likely to be correlated to stock price and performance versus social media sentiment. Stocksnips measures sentiment for relevant comments and statements about various business factors affecting a company’s performance from news articles and SEC filings. These relevant “snippets” are then analyzed to determine whether the snippet is a positive or negative comment.
Several studies have shown that news stories, especially around the company’s earnings announcement, influence stock price. Tetlock et al in their research paper in June 2008 concluded that linguistic media content captures otherwise hard-to-quantify aspects of firms’ fundamentals, which investors quickly incorporate into stock prices. It is known that stock price is influenced by numerous signals, including many internal and external factors which makes it difficult to predict stock price in general. To some extent a news based sentiment intrinsically covers a significant number of factors that affect stock price and therefore should be a more robust composite signal that should result in better correlation between stock price and stock sentiment.
Is Stock Price Correlated to StockSnips Sentiment?
StockSnips sentiment analysis is conducted by analyzing 50K+ news articles from 20+ sources every day. The relevant – performance related – comments made in the articles are analyzed and a proprietary machine learning model assesses the sentiment as either being positive or negative. Each such relevant comment is termed as a “snippet”. A 7 Day % positive sentiment is the ratio of positive news “snippets” to total snippets over the prior seven day period (including the current day).
We present the results of our analysis the correlation between stock price and StockSnips sentiment using daily StockSnips generated news sentiment data for Apple, Inc (ticker APPL) from April 7, 2016 to October 26, 2016.
We see a strong correlation between the 7 day moving average % positive sentiment and daily stock price. Notice the sharp dip in price after the 4/26/2016 earnings release. The news sentiment was negative prior to this date on news that Apple sales in China were declining. The sentiment was trending negative starting 4/11/2016. The stock price started declining soon after starting 4/16/2016. The sharp fall in sentiment in June was most likely driven by Brexit – the effect of which can be seen on Apple’s stock price also.
The above single stock analysis, shows a clear relationship between stock price and sentiment score with a correlation of 0.72 which is statistically significant at the 90% confidence level (p-value < 0.10). The larger question however is whether the sentiment signal is a leading indicator or not (since if not, or if lagging there is minimal value to using it for stock portfolio management).
Can StockSnips Sentiment be Used for Trading & Portfolio Management?
To address this question we assess stock price versus sentiment deviation (or change) from a baseline. The baseline sentiment is equivalent to a long-term trend of sentiment and the Sentiment Deviation is the difference between a 7 Day % positive sentiment minus the baseline sentiment (e.g. if 7Day is 85% and baseline is 55% - the deviation is +30%). The chart below shows Apple Stock Price versus Sentiment Deviation.
Over a period of time it is evident that a negative Sentiment Deviation is associated with weak stock price performance and vice-versa (as shown in red and green bars and the associated stock price level above). Furthermore, the sentiment deviation in many cases appears to precede stock price movements – which could indicate a leading effect:
1 . For item marked (1) – we can see prior to the Q1 Earnings release on 4/26 there is a marked flip in % 7 Day sentiment (versus baseline sentiment) which persisted for 6-7 days before the earnings, following which there is a steep descent in stock price.
2 . For item marked (2) – we can again see 7-10 days of negative sentiment deviation prior to a 5-10% drop in stock price (the whole red deviation block is off shifted to the left versus the stock decline).
3 . For item marked (3) – we can see strong positive deviation for a period of time (followed by an inexplicable negative deviation just prior to Q2 release). However, overall we can observe a strong positive growth from 7/7 all the way to 8/18.
4 . For item marked (4) – we can see strong and sustained positive deviation which again is followed by continued stock appreciation (again the whole “green positive deviation” is off shifted to right compared to the stock growth trajectory).
To analyze the strength and duration of the leading effect of StockSnips sentiment scores, Stock Price was correlated to Sentiment Deviation with various time lags as shown in the chart below.
In the above chart, current day stock price shows a 0.76 correlation to current day sentiment deviation. The correlation of stock price to sentiment deviation from 3 days ago is 0.74 and 5 days ago is 0.73, and from 10 days ago is 0.61. This potential “leading” effect of sentiment deviation on stock price is clearly indicated in this analysis implying that we could potentially predict the impact of current stock price using five day old sentiment deviation as shown in the chart below.
Ex-post stock price prediction (shown in red) is based on five day old sentiment data. As can be seen, while there is a persistent (2 to 3 dollar) difference in the level of predicted stock price, the model accurately predicts the downturn in Apple stock price over the prediction period – and the similarity is significant.
The statistically significant correlation of stock price with sentiment (and sentiment deviation) of 0.76 is a strong indicator of the importance of the StockSnips sentiment signal – at least for Apple stock. There are also other clear technical buy/sell signals that could be interpreted depending on signal strength and other factors. Additional research and modeling will need to be done across other companies to further validate the model. In addition, lagged market factors such as “total market sentiment average” and other signals could also be considered. Overall, the current research provides strong evidence that StockSnips sentiment scores could be used for portfolio management.
What is the Reliability of StockSnips Sentiment Score?
It is well known that stock price is impacted by endogenous factors such as earnings per share and growth, and exogenous factors including macro-economic factors (local and global economy) and “random white noise” events. Historically, and even today, accurate prediction of stock price is inherently complex if not impossible. However, in a globally connected world, companies are tied to other companies and large conglomerates (such as Apple) can impact the performance of other companies. If there is asymmetry in timing of information published by these interconnected companies, it is posited that such information can be useful for predicting likely stock movements. In the case of Apple, articles about the likely performance of Apple prior to earnings release were based on the performance results announced for related companies (example, Foxcon). The sentiment signal from such articles is a strong correlate to stock price and can be expected given the interdependence. However, in the case of smaller companies or more “insular” companies with fewer news articles, would sentiment be as strongly correlated to stock performance? It is unlikely, simply due to the fact that there would be fewer news articles for such stocks and the reliability of the sentiment signal would be questionable. Furthermore, information timing asymmetry wouldn’t exist and the leading indicator value of sentiment would also be questionable. Therefore it would appear that StockSnips sentiment signal would be more useful for large conglomerates with a large number of reliable news articles. Much more research is warranted is this regard and StockSnips is continuing to investigate additional companies and its sentiment algorithm with the goal of building a reliable sentiment based predictive model.