Reality Check: A Linguistic Analysis-Based Redefinition of Quarterly Earnings Surprise

Abstract

The media and financial community often rely on “earnings surprise” – the difference between company performance and the average of financial analysts’ forecasts – to summarize the manner in which earnings reports provide the market with new, unexpected information. Prior studies have found such a metric to exhibit a host of flaws, however, suggesting there may be a place for an alternative way to think about how quarterly earnings and investor expectations interact. In this study, I seek to shift the focus from numerical earnings figures to linguistic analyses of textual statements from both investors and management. I define an alternative metric called the “Reality Check” as the difference between the management sentiment as expressed on earnings conference calls and investor sentiment as expressed on Twitter – each of which has been found in prior studies to correlate with stock market moves. Using a sample of over 4,500 earnings reports from 490 companies, I test whether this metric can predict market reactions to earnings, finding it to have statistically significant predictive power for both short-term stock returns and stock price movements in the 15-day window before and after earnings. Despite this finding, the metric exhibits worse predictive power than traditional consensus-based earnings surprise when modeled independently and only adds a slight amount of predictive power when modeled together. This suggests that significant improvements would be required before such a metric could merit a place next to consensus-based forecasts when evaluating earnings reports.