Seeing Through the Spin: The Effect of News Sentiment on Firms' Stock Market Performance

We show that Stock market investors react only on the objective facts and not the spin in media articles. We use natural language processing tools to compute the tone of 288 thousands articles written by Reuters between 2000 and 2018, and show that it predicts the short-term stock market performance of companies. However, by exploiting a combination of unsupervised machine learning and econometric techniques, we show that this effect is only due to the informational content of the article, and not the framing of that article. The market sees through media spin and can filter informational content from irrelevant tone.

Abstract

The sentiment of news predicts the short-term stock market performance of individual companies. We find that this association is solely due to the idiosyncratic informational content of an article. We transparently quantify the association between news sentiment and stock market performance of S&P 500 companies, using articles written by Reuters between 2000 and 2018. First, we isolate the effect of sentiment independently of idiosyncratic informational content by exploiting a topic-based shift-share instrument. Second, we show that exogenous variation in article sentiment isolated through our topic-based shift-share instrument, while strongly related to article sentiment, is unrelated to abnormal returns in the stock market.