Effects of Annual Report Sentiment on Stock Returns

This study utilizes natural language processing techniques to analyze annual report narratives. The sentiment of annual report narratives is gauged by utilizing the frequency of words related to a linguistic category to establish sentiment. This sentiment is used to predict abnormal stock returns. We find that the indicators ”positivity”, ”constraining”, and ”superfluous” can, in a model, predict abnormal stock returns. When combined with interaction effects deriving from cultural differences, these models show an even stronger predicting capability for abnormal stock returns. When considering only uni-dimensional models, we find that all of the sentiment indicators have significant effects on abnormal stock returns in the short-term. Among the earlier mentioned indicators, these are: ”readability”, ”uncertainty”, ”litigious”, and ”text density”. These conclusions are drawn after controlling for year- or country-specific trends. Consequently, investors and companies alike are advised to analyze the sentiment of annual reports, as the information included in them is likely to cause short-term adaptations in the stock price.