When does the tone of earnings press releases matter?

Abstract

The tone of a firm's financial disclosure is increasingly used as a variable in panel data regressions to predict future performance and explain investors' reaction at earnings announcement. We investigate when tone is informative, and argue that the informativeness of tone increases with the information asymmetry between firms and investors. Using a sample of over 50,000 earnings press releases of about 1800 U.S. public firms between 2004 and 2015, we find that firm growth, size, age, complexity and forecast inaccuracy are key drivers of tone informativeness. The effect is economically significant, since, compared to the reference case of a transparent firm, we find that the slope coefficient of tone doubles or even quadruples in panel data regressions when the firm operates in an environment with high information asymmetry.

Disclosure tone management and labor unions

“…(2012) examine the information value of earnings press releases and document the positive correlations between the tone of earnings press releases and current or future firm profitability and between tone and investors’ reaction at the earnings announcement. Similarly, Boudt, Thewissen, and Torsin (2018) find that the tone of earnings press releases provides information value to shareholders and the informativeness of the tone increases with the level of information asymmetry. Overall, the literature supports the notion that the qualitative information in earnings releases signals a firm's financial situation.…”

Econometrics Meets Sentiment: An Overview of Methodology and Applications

“…Boudt et al . (2018) take the centered average of the scores coming from the lexicons they apply. A more statistical approach is commonality extraction, using principal component analysis or latent factor modeling.…”

“…Boudt et al . (2018) assess the interaction of sentiment with various company variables (finding that the informativeness of sentiment depends on the level of information asymmetry), while Arslan‐Ayaydin et al . (2016) interact sentiment with managerial compensation (finding that the informativeness of sentiment depends on the incentives to manipulate the sentiment).…”

“…Another simple scenario is to average the sentiment scores from any given number of methods to obtain an averaged sentiment score. Boudt et al (2018) take the centered average of the scores coming from the lexicons they apply. A more statistical approach is commonality extraction, using principal component analysis or latent factor modeling.…”

“…These interaction terms add information, allowing one, for example, to uncover that negative unusual news leads to an increase in U.S. stock market volatility (Glasserman and Mamaysky, 2019). Boudt et al (2018) assess the interaction of sentiment with various company variables (finding that the informativeness of sentiment depends on the level of information asymmetry), while Arslan-Ayaydin et al (2016) interact sentiment with managerial compensation (finding that the informativeness of sentiment depends on the incentives to manipulate the sentiment). García (2013) interacts a measure based on the New York Times news with a dummy variable to indicate a recession and concludes that daily stock returns are better predicted during recessions.…”