Sentiment Analysis for Stock Returns Using the TRNA Dataset

Abstract

In recent years there has been an increase in news related information for traded assets in the equity market. Financial news can be accessed in real time through online sources such as social media and Internet news. Traders may then take into account the news information when buying or selling equity stocks – which can drive stock prices. In this paper, I am analyzing whether a stock’s daily sentiment influences its daily return using a multi-factor model. I use the Thomson Reuters News Analytics (TRNA) data set, which gathers financial news stories in real time and converts them to sentiment scores. It uses a unique com- putation linguistic processing system to provide real-time numerical insight on financial events from multiple news sources. In this study, I use sentiment data from 21 big public companies and include the daily sentiment score of these com- panies along with lagged scores in the Fama-French three factor model to assess the influence of sentiment score on daily return. This is done using Ordinary Least Square (OLS) regression. My results suggest that even when taking into account the market factors, the daily sentiment score does significantly affect the same day stock returns for these 21 big public companies. However, the magni- tude of the effect is dependent on the time period, and some of the lag scores are significant as well. To test out the profitability of news sentiment signals, I sim- ulate a portfolio that longs high-positive sentiment securities in an out of sample period from 1/1/2019 to 3/14/2020. I also generated a profit and loss graph for the portfolio. Lastly, I extend my analysis by building a machine learning classifier model using the L1 regularized logistic regression to predict next-day return – classifying the next-day return as positive or negative. The result is below average, suggesting that while the daily sentiment score correlates with same-day return, it is not a good predictor for the next-day return.